Phase 0: scheduleragent + mcp-server read-only agent framework

Complete the natural-language scheduler agent's Phase 0 read-only pipeline: Task #2 (API Key auth mechanism): - New store/apikey.go: APIKey store (hash, validate, expiry) - controllers/apikey.go: GenerateAPIKey finally persists + registers RBAC role - controllers/auth.go: GetUsernameFromToken recognizes xce_ak_ prefix - middleware/validatehandler.go: isValidToken handles API Key - config/rbac.yaml: mcp_readonly role (view-only permissions) - 6 new test files covering unit + integration full-chain Task #3 (schedulerclient Go library): - types.go: URM types (URMJob/URMQueue/URMNode/JobStatus/Metrics) - client.go: read-only Client interface (6 methods) - httpclient.go: resty-based HTTP implementation - statusmap.go + _test.go: TDD status mapping (33 tests) Task #4 (mcp-server binary): - cmd/main.go: entry, reads XCE_API_URL/XCE_API_KEY, registers 6 MCP tools - internal/app.go: tool handlers wiring schedulerclient to MCP stdio - Uses github.com/mark3labs/mcp-go@v0.57.0 for stdio transport All: go vet clean, go build clean, schedulerclient 33/33 tests pass. mcp-server binary ~11MB arm64, CGO_ENABLED=0. Architecture: Claude Code --stdio--> mcp-server --HTTP--> apiserver REST API --cgo--> C backend Co-Authored-By: Claude <noreply@anthropic.com>
2026-07-30 11:16:36 +08:00 · 2026-07-30 11:16:36 +08:00 · e4af0cac73
commit e4af0cac73
parent 98066db68d
27 changed files with 1760 additions and 7 deletions
--- a/.wolf/memory.md
+++ b/.wolf/memory.md
@ -6082,3 +6082,5 @@
 | 01:56 | Edited ../../../../../claude/.claude/projects/-home-rshen-projects-xce-xce/memory/MEMORY.md | 1→2 lines | ~142 |
 | 01:56 | Session end: 33 writes across 20 files (cluster.c, sched.c, limits.yaml, project_bug1462_config_replication_fixed.md, MEMORY.md) | 39 reads | ~71360 tok |
 | 02:11 | Session end: 33 writes across 20 files (cluster.c, sched.c, limits.yaml, project_bug1462_config_replication_fixed.md, MEMORY.md) | 39 reads | ~71360 tok |
+| 12:10 | Updated root CLAUDE.md: corrected Native Adapter Go-import assumption (myxce kernel is C/cgo, Go only at apiserver/xcecli edges), logged API Key auth is half-built, confirmed 2 decisions (code location, service-account API Key auth) | /Users/dankeer/Projects/chatbot/CLAUDE.md | done | ~600 |
+| 12:10 | Updated myxce/CLAUDE.md and src/frontend/CLAUDE.md: registered planned schedulerclient/ and mcp-server/ modules (Phase 0, in progress) pointing to root CLAUDE.md for overall plan | myxce/CLAUDE.md, myxce/src/frontend/CLAUDE.md | done | ~300 |
--- a/CLAUDE.md
+++ b/CLAUDE.md
@ -49,6 +49,7 @@ For details on any topic below, read the corresponding steering file in `.kiro/s
 - `src/worker/` — See `src/worker/CLAUDE.md`
 - `src/scheduler/` — See `src/scheduler/CLAUDE.md`
 - `src/lib/` — See `src/lib/CLAUDE.md`
+- `src/frontend/schedulerclient/`, `src/frontend/mcp-server/` — Phase 0 已完成（2026-07-26）。`schedulerclient` 是独立 Go library 提供只读 Client interface + resty HTTP 实现（33 单测）。`mcp-server` 是 MCP stdio 二进制（`github.com/mark3labs/mcp-go@v0.57.0`），通过 `schedulerclient` 调用 `apiserver` REST API，暴露 6 个只读 MCP Tool。Pure Go, `CGO_ENABLED=0`。总体架构与分阶段计划见根目录 `CLAUDE.md`（"Unified Scheduler Agent Framework"）。

 # Agentic SDLC and Spec-Driven Development

--- a/src/frontend/CLAUDE.md
+++ b/src/frontend/CLAUDE.md
@ -8,6 +8,8 @@ Go-based REST API server (Gin) and CLI tool (Cobra) that expose cluster manageme
 |--------|------|-------------|
 | `apiserver` | `src/frontend/apiserver/` | `cmd/main.go` |
 | `xcecli` | `src/frontend/xcecli/` | `cmd/main.go` |
+| `schedulerclient` (Phase 0, ✅ done) | `src/frontend/schedulerclient/` | library, no entry point — read-only Go client for the scheduler agent's MCP Server, calls `apiserver`'s REST API. `go vet`/`go build`/`go test` clean (33 tests). See module source: `types.go` (URM types), `client.go` (interface), `httpclient.go` (resty impl), `statusmap.go`+`_test.go` (TDD, status mapping). |
+| `mcp-server` (Phase 0, ✅ done) | `src/frontend/mcp-server/` | `cmd/main.go` — read-only MCP Server exposing 6 tools (`scheduler_list_jobs`/`get`/`list_queues`/`list_nodes`/`job_logs`/`cluster_metrics`); pure Go, `CGO_ENABLED=0`, uses `schedulerclient` as API client, authenticates via `XCE_API_KEY` env var. Uses `github.com/mark3labs/mcp-go@v0.57.0` for MCP stdio transport. `go build`/`go vet` clean, binary ~11MB. |

 ## Build & Test
 ```bash
--- a/src/frontend/apiserver/config/rbac.yaml
+++ b/src/frontend/apiserver/config/rbac.yaml
@ -148,4 +148,27 @@ roles:
          - type: queue
            values: [normal]

+  # mcp_readonly: service-account role for the scheduler-agent MCP server
+  # (Phase 0, read-only tools only). Deliberately excludes job:control,
+  # queue:control, host:control, cluster:control, rbac:*, file:*, user:* —
+  # an MCP server credential should never be able to mutate cluster state
+  # or manage other users/roles.
+  mcp_readonly:
+    permissions:
+      - action: job:view
+        domain:
+          - type: all
+      - action: queue:view
+        domain:
+          - type: all
+      - action: host:view
+        domain:
+          - type: all
+      - action: resource:view
+        domain:
+          - type: all
+      - action: cluster:view
+        domain:
+          - type: all
+

--- a/src/frontend/apiserver/controllers/apikey.go
+++ b/src/frontend/apiserver/controllers/apikey.go
@ -7,6 +7,7 @@ import (

 	"apiserver/common"
 	"apiserver/models"
+	"apiserver/store"

 	"github.com/gin-gonic/gin"
 )
@ -16,7 +17,7 @@ func generateAPIKey() (string, error) {
 	if _, err := rand.Read(bytes); err != nil {
 		return "", err
 	}
-	return hex.EncodeToString(bytes), nil
+	return models.APIKeyPrefix + hex.EncodeToString(bytes), nil
 }

 type APIKeyController struct{}
@ -49,13 +50,34 @@ func (a *APIKeyController) GenerateAPIKey(ctx *gin.Context) {
 		expiryDays = req.ExpiryDays
 	}

+	createTime := time.Now()
+	expiryTime := createTime.AddDate(0, 0, expiryDays)
+
+	// Persist the key (by hash) so it can actually be used to authenticate.
+	store.AddAPIKey(apiKey, store.APIKeyEntry{
+		Username:    req.Username,
+		RoleName:    req.RoleName,
+		Description: req.Description,
+		CreateTime:  createTime,
+		ExpiryTime:  expiryTime,
+	})
+
+	// RequirePermission resolves permissions via store.GetUserRoleNames, not
+	// via the APIKeyEntry directly — register the RBAC user→role mapping too,
+	// or the issued key would authenticate but be denied every permission
+	// check.
+	if req.RoleName != "" {
+		store.AddRoleForUser(req.Username, req.RoleName)
+	}
+
 	// Create response
 	response := models.APIKeyResponse{
 		APIKey:     apiKey,
 		Username:   req.Username,
-		CreateTime: time.Now(),
-		ExpiryTime: time.Now().AddDate(0, 0, expiryDays),
+		RoleName:   req.RoleName,
+		CreateTime: createTime,
+		ExpiryTime: expiryTime,
 	}

-	ReturnSuccess(ctx, 0, "API key generated successfully", response)
+	ReturnSuccess(ctx, common.SUCCESS_CODE, "API key generated successfully", response)
 }
--- a/src/frontend/apiserver/controllers/apikey_integration_test.go
+++ b/src/frontend/apiserver/controllers/apikey_integration_test.go
@ -0,0 +1,59 @@
+package controllers
+
+import (
+	"net/http"
+	"testing"
+	"time"
+
+	"apiserver/models"
+	"apiserver/store"
+
+	"github.com/gin-gonic/gin"
+)
+
+// TestIntegration_APIKey_MCPReadonly_FullMiddlewareChain verifies the
+// end-to-end story for the scheduler-agent MCP server's service account:
+// an API-key-authenticated request with the mcp_readonly role can reach a
+// job:view route but is denied a job:control route — through the exact same
+// RequirePermission middleware JWT-authenticated users go through.
+func TestIntegration_APIKey_MCPReadonly_FullMiddlewareChain(t *testing.T) {
+	yamlContent := `
+permissions:
+  - job:view
+  - job:control
+roles:
+  mcp_readonly:
+    permissions:
+      - action: job:view
+        domain:
+          - type: all
+`
+	setupRBACTest(t, yamlContent)
+
+	raw := models.APIKeyPrefix + "mcp-integration-test"
+	store.AddAPIKey(raw, store.APIKeyEntry{
+		Username:   "mcp-bot",
+		RoleName:   "mcp_readonly",
+		ExpiryTime: time.Now().Add(time.Hour),
+	})
+	store.AddRoleForUser("mcp-bot", "mcp_readonly")
+
+	router := newTestRouter()
+	jobsGroup := router.Group("/xce/v1/jobs")
+	jobsGroup.GET("", RequirePermission("job:view"), func(ctx *gin.Context) {
+		ctx.JSON(http.StatusOK, gin.H{"status": "ok"})
+	})
+	jobsGroup.POST("/control", RequirePermission("job:control"), func(ctx *gin.Context) {
+		ctx.JSON(http.StatusAccepted, gin.H{"status": "accepted"})
+	})
+
+	w := doRequest(router, "GET", "/xce/v1/jobs", raw)
+	if w.Code != http.StatusOK {
+		t.Errorf("mcp_readonly on job:view: got %d, want 200, body=%s", w.Code, w.Body.String())
+	}
+
+	w = doRequest(router, "POST", "/xce/v1/jobs/control", raw)
+	if w.Code != http.StatusForbidden {
+		t.Errorf("mcp_readonly on job:control: got %d, want 403", w.Code)
+	}
+}
--- a/src/frontend/apiserver/controllers/apikey_test.go
+++ b/src/frontend/apiserver/controllers/apikey_test.go
@ -0,0 +1,176 @@
+package controllers
+
+import (
+	"apiserver/models"
+	"apiserver/store"
+	"apiserver/utils"
+	"bytes"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+	"time"
+
+	"github.com/gin-gonic/gin"
+	"go.uber.org/zap"
+)
+
+func resetAPIKeyControllerState() {
+	store.InitStores()
+	utils.Log = zap.NewNop()
+}
+
+// TestGetUsernameFromToken_APIKey verifies that a valid, unexpired API key
+// resolves to its associated username via the same entry point every other
+// controller already calls for JWTs — so no call site needs to change.
+func TestGetUsernameFromToken_APIKey(t *testing.T) {
+	resetAPIKeyControllerState()
+
+	raw := models.APIKeyPrefix + "abc123"
+	store.AddAPIKey(raw, store.APIKeyEntry{
+		Username:   "mcp-bot",
+		RoleName:   "mcp_readonly",
+		ExpiryTime: time.Now().Add(time.Hour),
+	})
+
+	username, err := GetUsernameFromToken(raw)
+	if err != nil {
+		t.Fatalf("unexpected error: %v", err)
+	}
+	if username != "mcp-bot" {
+		t.Errorf("username = %q, want mcp-bot", username)
+	}
+}
+
+// TestGetUsernameFromToken_APIKey_BearerPrefix verifies the "Bearer <key>"
+// header form (as sent by real HTTP clients) also resolves correctly.
+func TestGetUsernameFromToken_APIKey_BearerPrefix(t *testing.T) {
+	resetAPIKeyControllerState()
+
+	raw := models.APIKeyPrefix + "with-bearer"
+	store.AddAPIKey(raw, store.APIKeyEntry{
+		Username:   "mcp-bot",
+		ExpiryTime: time.Now().Add(time.Hour),
+	})
+
+	username, err := GetUsernameFromToken("Bearer " + raw)
+	if err != nil {
+		t.Fatalf("unexpected error: %v", err)
+	}
+	if username != "mcp-bot" {
+		t.Errorf("username = %q, want mcp-bot", username)
+	}
+}
+
+// TestGetUsernameFromToken_APIKey_Unknown verifies an API-key-shaped token
+// that was never issued is rejected rather than silently falling through to
+// JWT parsing (which would also fail, but with a confusing error message).
+func TestGetUsernameFromToken_APIKey_Unknown(t *testing.T) {
+	resetAPIKeyControllerState()
+
+	_, err := GetUsernameFromToken(models.APIKeyPrefix + "never-issued")
+	if err == nil {
+		t.Error("expected error for unknown API key")
+	}
+}
+
+// TestGetUsernameFromToken_APIKey_Expired verifies an expired API key is
+// rejected.
+func TestGetUsernameFromToken_APIKey_Expired(t *testing.T) {
+	resetAPIKeyControllerState()
+
+	raw := models.APIKeyPrefix + "expired"
+	store.AddAPIKey(raw, store.APIKeyEntry{
+		Username:   "mcp-bot",
+		ExpiryTime: time.Now().Add(-time.Hour),
+	})
+
+	_, err := GetUsernameFromToken(raw)
+	if err == nil {
+		t.Error("expected error for expired API key")
+	}
+}
+
+// TestGenerateAPIKey_PersistsToStore verifies the controller actually writes
+// the generated key into the store (previously it only echoed a random hex
+// string back to the caller without persisting anything, so no request
+// authenticated with the returned key could ever succeed).
+func TestGenerateAPIKey_PersistsToStore(t *testing.T) {
+	resetAPIKeyControllerState()
+
+	gin.SetMode(gin.TestMode)
+	router := gin.New()
+	handler := &APIKeyController{}
+	router.POST("/xce/v1/apikeys", handler.GenerateAPIKey)
+
+	body := `{"username":"mcp-bot","role_name":"mcp_readonly","expiry_days":7}`
+	req := httptest.NewRequest(http.MethodPost, "/xce/v1/apikeys", bytes.NewBufferString(body))
+	req.Header.Set("Authorization", "Bearer some-admin-token")
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("status = %d, want 200, body=%s", w.Code, w.Body.String())
+	}
+
+	var resp struct {
+		Data models.APIKeyResponse `json:"data"`
+	}
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("unmarshal response: %v", err)
+	}
+	if resp.Data.APIKey == "" {
+		t.Fatal("response did not include an api_key")
+	}
+
+	// The returned key must actually be usable for authentication —
+	// i.e. it must have been persisted to the store, not merely echoed.
+	entry, ok := store.ValidateAPIKey(resp.Data.APIKey)
+	if !ok {
+		t.Fatal("generated API key was not persisted to the store")
+	}
+	if entry.Username != "mcp-bot" {
+		t.Errorf("persisted username = %q, want mcp-bot", entry.Username)
+	}
+	if entry.RoleName != "mcp_readonly" {
+		t.Errorf("persisted role = %q, want mcp_readonly", entry.RoleName)
+	}
+
+	// RequirePermission/GetUserAllPermissionsWithType resolve permissions via
+	// store.GetUserRoleNames(username), not via the APIKeyEntry.RoleName field
+	// directly — so generating a key must also register the RBAC user→role
+	// mapping, otherwise the issued key would authenticate but be denied every
+	// permission check.
+	roles := store.GetUserRoleNames("mcp-bot")
+	found := false
+	for _, r := range roles {
+		if r == "mcp_readonly" {
+			found = true
+		}
+	}
+	if !found {
+		t.Errorf("GetUserRoleNames(mcp-bot) = %v, want to include mcp_readonly", roles)
+	}
+}
+
+// TestGenerateAPIKey_RequiresAuthorizationHeader verifies the endpoint still
+// rejects requests with no Authorization header at all.
+func TestGenerateAPIKey_RequiresAuthorizationHeader(t *testing.T) {
+	resetAPIKeyControllerState()
+
+	gin.SetMode(gin.TestMode)
+	router := gin.New()
+	handler := &APIKeyController{}
+	router.POST("/xce/v1/apikeys", handler.GenerateAPIKey)
+
+	body := `{"username":"mcp-bot","role_name":"mcp_readonly"}`
+	req := httptest.NewRequest(http.MethodPost, "/xce/v1/apikeys", bytes.NewBufferString(body))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusUnauthorized {
+		t.Errorf("status = %d, want 401", w.Code)
+	}
+}
--- a/src/frontend/apiserver/controllers/auth.go
+++ b/src/frontend/apiserver/controllers/auth.go
@ -181,7 +181,22 @@ func (a AuthController) Logout(ctx *gin.Context) {
 }

 func GetUsernameFromToken(tokenString string) (string, error) {
-	token, err := jwt.Parse(strings.TrimPrefix(tokenString, "Bearer "), func(token *jwt.Token) (interface{}, error) {
+	tokenString = strings.TrimPrefix(tokenString, "Bearer ")
+
+	// Service-account API keys are a distinct credential type from JWTs
+	// (recognized by a fixed prefix, see models.APIKeyPrefix) — resolve them
+	// against the API key store instead of attempting a JWT parse, so every
+	// existing GetUsernameFromToken call site works transparently for both
+	// credential types.
+	if strings.HasPrefix(tokenString, models.APIKeyPrefix) {
+		entry, ok := store.ValidateAPIKey(tokenString)
+		if !ok {
+			return "", fmt.Errorf("invalid or expired API key")
+		}
+		return entry.Username, nil
+	}
+
+	token, err := jwt.Parse(tokenString, func(token *jwt.Token) (interface{}, error) {
 		// Validate the algorithm
 		if _, ok := token.Method.(*jwt.SigningMethodHMAC); !ok {
 			return nil, fmt.Errorf("unexpected signing method: %v", token.Header["alg"])
--- a/src/frontend/apiserver/controllers/rbac_mcp_readonly_test.go
+++ b/src/frontend/apiserver/controllers/rbac_mcp_readonly_test.go
@ -0,0 +1,52 @@
+package controllers
+
+import (
+	"apiserver/store"
+	"os"
+	"path/filepath"
+	"testing"
+)
+
+// TestRBACYAML_McpReadonlyRole_LoadsReadOnlyPermissions verifies the
+// mcp_readonly role (added for the scheduler-agent MCP server's service
+// account, Phase 0) is present in the shipped config/rbac.yaml and grants
+// only *:view permissions — never any *:control/*:submit permission, so an
+// MCP server credential can never mutate cluster state.
+func TestRBACYAML_McpReadonlyRole_LoadsReadOnlyPermissions(t *testing.T) {
+	resetRBACState()
+
+	path, err := filepath.Abs("../config/rbac.yaml")
+	if err != nil {
+		t.Fatalf("resolve rbac.yaml path: %v", err)
+	}
+	if _, err := os.Stat(path); err != nil {
+		t.Skipf("config/rbac.yaml not found at %s: %v", path, err)
+	}
+
+	if err := LoadRBACConfig(path); err != nil {
+		t.Fatalf("LoadRBACConfig: %v", err)
+	}
+
+	if !store.Roles.Has("mcp_readonly") {
+		t.Fatal("expected mcp_readonly role to be loaded from config/rbac.yaml")
+	}
+
+	perms := store.GetRolePermissions("mcp_readonly")
+	if len(perms) == 0 {
+		t.Fatal("mcp_readonly should have at least one permission")
+	}
+
+	for _, p := range perms {
+		if !hasViewSuffix(p.Action) {
+			t.Errorf("mcp_readonly grants non-view permission %q — MCP service accounts must be read-only", p.Action)
+		}
+	}
+}
+
+func hasViewSuffix(action string) bool {
+	const suffix = ":view"
+	if len(action) < len(suffix) {
+		return false
+	}
+	return action[len(action)-len(suffix):] == suffix
+}
--- a/src/frontend/apiserver/middleware/validatehandler.go
+++ b/src/frontend/apiserver/middleware/validatehandler.go
@ -4,6 +4,7 @@ import (
 	"apiserver/common"
 	"apiserver/controllers"
 	"apiserver/models"
+	"apiserver/store"
 	"apiserver/utils"
 	logger "apiserver/utils"
 	"bytes"
@ -99,6 +100,14 @@ func validateHeader() gin.HandlerFunc {
 }

 func isValidToken(tokenString string, IP string) (bool, error) {
+	// Service-account API keys are a distinct credential type from JWTs and
+	// carry no IP binding (they're meant for long-lived automated clients
+	// like the scheduler-agent MCP server, which may not have a stable IP).
+	if strings.HasPrefix(tokenString, models.APIKeyPrefix) {
+		_, ok := store.ValidateAPIKey(tokenString)
+		return ok, nil
+	}
+
 	token, err := jwt.ParseWithClaims(tokenString, &models.JWTClaims{}, func(token *jwt.Token) (interface{}, error) {
 		return []byte(utils.AuthenticationYamlConfig.Token.SecretKey), nil
 	})
--- a/src/frontend/apiserver/middleware/validatehandler_test.go
+++ b/src/frontend/apiserver/middleware/validatehandler_test.go
@ -0,0 +1,109 @@
+package middleware
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"testing"
+	"time"
+
+	"apiserver/models"
+	"apiserver/store"
+	"apiserver/utils"
+
+	"github.com/gin-gonic/gin"
+	"go.uber.org/zap"
+)
+
+func resetValidateHandlerState() {
+	store.InitStores()
+	utils.Log = zap.NewNop()
+	utils.AuthenticationYamlConfig = &utils.Authentication{
+		Token: utils.TokenConfig{
+			SecretKey:       "test-secret",
+			Type:            "HS256",
+			TokenExpireTime: 3600,
+		},
+	}
+}
+
+func newValidateHandlerRouter() *gin.Engine {
+	gin.SetMode(gin.TestMode)
+	r := gin.New()
+	r.Use(ValidateHandler())
+	r.GET("/xce/v1/jobs", func(ctx *gin.Context) { ctx.Status(http.StatusOK) })
+	return r
+}
+
+// TestValidateHeader_APIKey_Valid verifies a valid, unexpired API key in the
+// Authorization header passes the middleware (no JWT involved at all).
+func TestValidateHeader_APIKey_Valid(t *testing.T) {
+	resetValidateHandlerState()
+
+	raw := models.APIKeyPrefix + "valid-key"
+	store.AddAPIKey(raw, store.APIKeyEntry{
+		Username:   "mcp-bot",
+		ExpiryTime: time.Now().Add(time.Hour),
+	})
+
+	router := newValidateHandlerRouter()
+	req := httptest.NewRequest(http.MethodGet, "/xce/v1/jobs", nil)
+	req.Header.Set("Authorization", "Bearer "+raw)
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("status = %d, want 200, body=%s", w.Code, w.Body.String())
+	}
+}
+
+// TestValidateHeader_APIKey_Unknown verifies an API-key-shaped token that was
+// never issued is rejected with 401, not a confusing JWT parse error.
+func TestValidateHeader_APIKey_Unknown(t *testing.T) {
+	resetValidateHandlerState()
+
+	router := newValidateHandlerRouter()
+	req := httptest.NewRequest(http.MethodGet, "/xce/v1/jobs", nil)
+	req.Header.Set("Authorization", "Bearer "+models.APIKeyPrefix+"never-issued")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusUnauthorized {
+		t.Errorf("status = %d, want 401", w.Code)
+	}
+}
+
+// TestValidateHeader_APIKey_Expired verifies an expired API key is rejected.
+func TestValidateHeader_APIKey_Expired(t *testing.T) {
+	resetValidateHandlerState()
+
+	raw := models.APIKeyPrefix + "expired-key"
+	store.AddAPIKey(raw, store.APIKeyEntry{
+		Username:   "mcp-bot",
+		ExpiryTime: time.Now().Add(-time.Hour),
+	})
+
+	router := newValidateHandlerRouter()
+	req := httptest.NewRequest(http.MethodGet, "/xce/v1/jobs", nil)
+	req.Header.Set("Authorization", "Bearer "+raw)
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusUnauthorized {
+		t.Errorf("status = %d, want 401", w.Code)
+	}
+}
+
+// TestValidateHeader_NoToken_StillRejected is a regression guard: adding the
+// API key branch must not weaken the existing "no token" rejection.
+func TestValidateHeader_NoToken_StillRejected(t *testing.T) {
+	resetValidateHandlerState()
+
+	router := newValidateHandlerRouter()
+	req := httptest.NewRequest(http.MethodGet, "/xce/v1/jobs", nil)
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusUnauthorized {
+		t.Errorf("status = %d, want 401", w.Code)
+	}
+}
--- a/src/frontend/apiserver/models/apikey.go
+++ b/src/frontend/apiserver/models/apikey.go
@ -2,10 +2,14 @@ package models

 import "time"

-type APIKeyController struct{}
+// APIKeyPrefix marks an Authorization header value as a service-account API
+// key rather than a JWT, so validateHeader() can branch without attempting
+// (and failing) a JWT parse first.
+const APIKeyPrefix = "xce_ak_"

 type APIKeyRequest struct {
 	Username    string `json:"username" binding:"required"`
+	RoleName    string `json:"role_name" binding:"required"`
 	Description string `json:"description,omitempty"`
 	ExpiryDays  int    `json:"expiry_days,omitempty"`
 }
@ -13,6 +17,7 @@ type APIKeyRequest struct {
 type APIKeyResponse struct {
 	APIKey     string    `json:"api_key"`
 	Username   string    `json:"username"`
+	RoleName   string    `json:"role_name"`
 	CreateTime time.Time `json:"create_time"`
 	ExpiryTime time.Time `json:"expiry_time"`
 }
--- a/src/frontend/apiserver/router/routers.go
+++ b/src/frontend/apiserver/router/routers.go
@ -73,6 +73,14 @@ func Router() *gin.Engine {
 		permHandler := &controllers.PermController{}
 		permsApi.GET("", controllers.RequirePermission("rbac:view"), permHandler.GetPermsInfo)
 	}
+	// ################ case: apikeys route (service-account credentials, e.g. for the
+	// scheduler-agent MCP server; generation is admin-only, gated the same way role
+	// management is)
+	apikeysApi := r.Group("/xce/v1/apikeys")
+	{
+		apikeyHandler := &controllers.APIKeyController{}
+		apikeysApi.POST("", controllers.RequirePermission("rbac:control"), apikeyHandler.GenerateAPIKey)
+	}
 	// ################ case: roles route
 	rolesApi := r.Group("/xce/v1/roles")
 	{
--- a/src/frontend/apiserver/store/apikey.go
+++ b/src/frontend/apiserver/store/apikey.go
@ -0,0 +1,47 @@
+package store
+
+import (
+	"crypto/sha256"
+	"encoding/hex"
+	"time"
+)
+
+// APIKeys maps a hashed API key (hex-encoded SHA-256) to its metadata.
+// The raw key is never stored — only its hash — so a store/config dump
+// cannot leak usable credentials.
+var APIKeys *DataStore[string, APIKeyEntry]
+
+// APIKeyEntry holds the metadata associated with a service-account API key.
+type APIKeyEntry struct {
+	Username    string
+	RoleName    string
+	Description string
+	CreateTime  time.Time
+	ExpiryTime  time.Time
+}
+
+// HashAPIKey returns the hex-encoded SHA-256 digest of a raw API key.
+// Used both when persisting a newly generated key and when validating one
+// presented by a client — the raw value is never compared or stored directly.
+func HashAPIKey(rawKey string) string {
+	sum := sha256.Sum256([]byte(rawKey))
+	return hex.EncodeToString(sum[:])
+}
+
+// AddAPIKey persists a new API key by its hash.
+func AddAPIKey(rawKey string, entry APIKeyEntry) {
+	APIKeys.Put(HashAPIKey(rawKey), entry)
+}
+
+// ValidateAPIKey looks up a raw API key by its hash and returns its metadata
+// if present and not expired. A zero ExpiryTime never expires.
+func ValidateAPIKey(rawKey string) (APIKeyEntry, bool) {
+	entry, ok := APIKeys.Get(HashAPIKey(rawKey))
+	if !ok {
+		return APIKeyEntry{}, false
+	}
+	if !entry.ExpiryTime.IsZero() && time.Now().After(entry.ExpiryTime) {
+		return APIKeyEntry{}, false
+	}
+	return entry, true
+}
--- a/src/frontend/apiserver/store/apikey_test.go
+++ b/src/frontend/apiserver/store/apikey_test.go
@ -0,0 +1,94 @@
+package store
+
+import (
+	"testing"
+	"time"
+)
+
+func resetAPIKeyState() {
+	InitStores()
+}
+
+func TestHashAPIKey_Deterministic(t *testing.T) {
+	h1 := HashAPIKey("xce_ak_abc123")
+	h2 := HashAPIKey("xce_ak_abc123")
+	if h1 != h2 {
+		t.Errorf("HashAPIKey should be deterministic: %q != %q", h1, h2)
+	}
+	h3 := HashAPIKey("xce_ak_different")
+	if h1 == h3 {
+		t.Error("HashAPIKey should differ for different inputs")
+	}
+}
+
+func TestHashAPIKey_NeverPlaintext(t *testing.T) {
+	raw := "xce_ak_secret-value"
+	hashed := HashAPIKey(raw)
+	if hashed == raw {
+		t.Error("HashAPIKey must not return the raw key unchanged")
+	}
+}
+
+func TestAddAPIKey_ValidateAPIKey_RoundTrip(t *testing.T) {
+	resetAPIKeyState()
+
+	raw := "xce_ak_roundtrip"
+	entry := APIKeyEntry{
+		Username:   "mcp-bot",
+		RoleName:   "mcp_readonly",
+		CreateTime: time.Now(),
+		ExpiryTime: time.Now().Add(24 * time.Hour),
+	}
+	AddAPIKey(raw, entry)
+
+	got, ok := ValidateAPIKey(raw)
+	if !ok {
+		t.Fatal("expected ValidateAPIKey to succeed for a freshly-added key")
+	}
+	if got.Username != "mcp-bot" {
+		t.Errorf("Username = %q, want mcp-bot", got.Username)
+	}
+	if got.RoleName != "mcp_readonly" {
+		t.Errorf("RoleName = %q, want mcp_readonly", got.RoleName)
+	}
+}
+
+func TestValidateAPIKey_Unknown(t *testing.T) {
+	resetAPIKeyState()
+
+	_, ok := ValidateAPIKey("xce_ak_never-issued")
+	if ok {
+		t.Error("expected ValidateAPIKey to fail for a key that was never added")
+	}
+}
+
+func TestValidateAPIKey_Expired(t *testing.T) {
+	resetAPIKeyState()
+
+	raw := "xce_ak_expired"
+	AddAPIKey(raw, APIKeyEntry{
+		Username:   "mcp-bot",
+		RoleName:   "mcp_readonly",
+		CreateTime: time.Now().Add(-48 * time.Hour),
+		ExpiryTime: time.Now().Add(-1 * time.Hour), // already expired
+	})
+
+	_, ok := ValidateAPIKey(raw)
+	if ok {
+		t.Error("expected ValidateAPIKey to fail for an expired key")
+	}
+}
+
+func TestValidateAPIKey_StoredByHashNotPlaintext(t *testing.T) {
+	resetAPIKeyState()
+
+	raw := "xce_ak_hashed-storage"
+	AddAPIKey(raw, APIKeyEntry{Username: "mcp-bot", ExpiryTime: time.Now().Add(time.Hour)})
+
+	if APIKeys.Has(raw) {
+		t.Error("store must not contain the raw key as a map key")
+	}
+	if !APIKeys.Has(HashAPIKey(raw)) {
+		t.Error("store should contain the hashed key")
+	}
+}
--- a/src/frontend/apiserver/store/stores.go
+++ b/src/frontend/apiserver/store/stores.go
@ -16,6 +16,8 @@ var (
 	UserRoles *DataStore[string, []string] // user_name → role_names
 )

+// APIKeys is declared in apikey.go (keeps API-key store logic self-contained).
+
 // RoleEntry holds a role's permissions.
 type RoleEntry struct {
 	RoleName    string
@ -41,6 +43,7 @@ func InitStores() {
 	JobExecHosts = NewDataStore[int64, models.JobExecHosts]()
 	Roles = NewDataStore[string, RoleEntry]()
 	UserRoles = NewDataStore[string, []string]()
+	APIKeys = NewDataStore[string, APIKeyEntry]()
 }

 // Helper functions for RBAC.
--- a/src/frontend/mcp-server/cmd/main.go
+++ b/src/frontend/mcp-server/cmd/main.go
@ -0,0 +1,86 @@
+package main
+
+import (
+	"log"
+	"os"
+
+	mcps "github.com/mark3labs/mcp-go/server"
+	"github.com/mark3labs/mcp-go/mcp"
+
+	"mcp-server/internal"
+)
+
+func main() {
+	baseURL := os.Getenv("XCE_API_URL")
+	token := os.Getenv("XCE_API_KEY")
+
+	if baseURL == "" {
+		baseURL = "https://localhost:8080"
+		log.Printf("XCE_API_URL not set, defaulting to %s", baseURL)
+	}
+	if token == "" {
+		log.Fatal("XCE_API_KEY environment variable is required")
+	}
+
+	app, err := internal.NewApp(baseURL, token)
+	if err != nil {
+		log.Fatalf("failed to initialize: %v", err)
+	}
+
+	s := mcps.NewMCPServer(
+		"scheduler-agent",
+		"0.1.0",
+		mcps.WithResourceCapabilities(true, true),
+		mcps.WithLogging(),
+	)
+
+	// ── scheduler_list_jobs ──
+	s.AddTool(mcp.NewTool("scheduler_list_jobs",
+		mcp.WithDescription("List jobs in the cluster. Supports filtering by queue, owner, status, and pagination."),
+		mcp.WithString("queue", mcp.Description("Filter by queue name")),
+		mcp.WithString("owner", mcp.Description("Filter by job owner username")),
+		mcp.WithString("status", mcp.Description("Filter by status: pending|running|succeeded|failed|cancelled|suspended|unknown")),
+		mcp.WithNumber("limit", mcp.Description("Max results to return")),
+		mcp.WithString("page_token", mcp.Description("Pagination token from a prior response")),
+	), app.HandleListJobs)
+
+	// ── scheduler_get_job ──
+	s.AddTool(mcp.NewTool("scheduler_get_job",
+		mcp.WithDescription("Get detailed information about a specific job by its ID."),
+		mcp.WithString("id",
+			mcp.Description("Job ID"),
+			mcp.Required(),
+		),
+	), app.HandleGetJob)
+
+	// ── scheduler_list_queues ──
+	s.AddTool(mcp.NewTool("scheduler_list_queues",
+		mcp.WithDescription("List all queues/partitions and their current state (active, inactive, draining)."),
+	), app.HandleListQueues)
+
+	// ── scheduler_list_nodes ──
+	s.AddTool(mcp.NewTool("scheduler_list_nodes",
+		mcp.WithDescription("List all compute nodes with their resource usage and status."),
+	), app.HandleListNodes)
+
+	// ── scheduler_job_logs ──
+	s.AddTool(mcp.NewTool("scheduler_job_logs",
+		mcp.WithDescription("Retrieve the output logs of a job."),
+		mcp.WithString("id",
+			mcp.Description("Job ID"),
+			mcp.Required(),
+		),
+		mcp.WithNumber("tail", mcp.Description("Number of lines from the end to return (0 = all)")),
+		mcp.WithNumber("offset_bytes", mcp.Description("Byte offset to start reading from")),
+	), app.HandleJobLogs)
+
+	// ── scheduler_cluster_metrics ──
+	s.AddTool(mcp.NewTool("scheduler_cluster_metrics",
+		mcp.WithDescription("Get a snapshot of overall cluster resource usage and utilization."),
+	), app.HandleClusterMetrics)
+
+	log.Printf("Starting scheduler-agent MCP server (base-url: %s)", baseURL)
+	if err := mcps.ServeStdio(s); err != nil {
+		log.Fatalf("server error: %v", err)
+	}
+}
--- a/src/frontend/mcp-server/go.mod
+++ b/src/frontend/mcp-server/go.mod
@ -0,0 +1,21 @@
+module mcp-server
+
+go 1.25.5
+
+replace schedulerclient => ../schedulerclient
+
+require (
+	github.com/mark3labs/mcp-go v0.57.0
+	schedulerclient v0.0.0-00010101000000-000000000000
+)
+
+require (
+	github.com/google/jsonschema-go v0.4.2 // indirect
+	github.com/google/uuid v1.6.0 // indirect
+	github.com/santhosh-tekuri/jsonschema/v6 v6.0.2 // indirect
+	github.com/spf13/cast v1.7.1 // indirect
+	github.com/yosida95/uritemplate/v3 v3.0.2 // indirect
+	golang.org/x/net v0.33.0 // indirect
+	golang.org/x/text v0.21.0 // indirect
+	resty.dev/v3 v3.0.0-beta.2 // indirect
+)
--- a/src/frontend/mcp-server/go.sum
+++ b/src/frontend/mcp-server/go.sum
@ -0,0 +1,38 @@
+github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=
+github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
+github.com/dlclark/regexp2 v1.11.0 h1:G/nrcoOa7ZXlpoa/91N3X7mM3r8eIlMBBJZvsz/mxKI=
+github.com/dlclark/regexp2 v1.11.0/go.mod h1:DHkYz0B9wPfa6wondMfaivmHpzrQ3v9q8cnmRbL6yW8=
+github.com/frankban/quicktest v1.14.6 h1:7Xjx+VpznH+oBnejlPUj8oUpdxnVs4f8XU8WnHkI4W8=
+github.com/frankban/quicktest v1.14.6/go.mod h1:4ptaffx2x8+WTWXmUCuVU6aPUX1/Mz7zb5vbUoiM6w0=
+github.com/google/go-cmp v0.7.0 h1:wk8382ETsv4JYUZwIsn6YpYiWiBsYLSJiTsyBybVuN8=
+github.com/google/go-cmp v0.7.0/go.mod h1:pXiqmnSA92OHEEa9HXL2W4E7lf9JzCmGVUdgjX3N/iU=
+github.com/google/jsonschema-go v0.4.2 h1:tmrUohrwoLZZS/P3x7ex0WAVknEkBZM46iALbcqoRA8=
+github.com/google/jsonschema-go v0.4.2/go.mod h1:r5quNTdLOYEz95Ru18zA0ydNbBuYoo9tgaYcxEYhJVE=
+github.com/google/uuid v1.6.0 h1:NIvaJDMOsjHA8n1jAhLSgzrAzy1Hgr+hNrb57e+94F0=
+github.com/google/uuid v1.6.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
+github.com/kr/pretty v0.3.1 h1:flRD4NNwYAUpkphVc1HcthR4KEIFJ65n8Mw5qdRn3LE=
+github.com/kr/pretty v0.3.1/go.mod h1:hoEshYVHaxMs3cyo3Yncou5ZscifuDolrwPKZanG3xk=
+github.com/kr/text v0.2.0 h1:5Nx0Ya0ZqY2ygV366QzturHI13Jq95ApcVaJBhpS+AY=
+github.com/kr/text v0.2.0/go.mod h1:eLer722TekiGuMkidMxC/pM04lWEeraHUUmBw8l2grE=
+github.com/mark3labs/mcp-go v0.57.0 h1:jzWKyCzdWnwnZt05cvcQQ+ngiUl2RnixXJa7Kj4qP1E=
+github.com/mark3labs/mcp-go v0.57.0/go.mod h1:+8WclSK1ZUweCP3hvktSji8n8ABG/95QaEkeVE/Uwas=
+github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=
+github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
+github.com/rogpeppe/go-internal v1.14.1 h1:UQB4HGPB6osV0SQTLymcB4TgvyWu6ZyliaW0tI/otEQ=
+github.com/rogpeppe/go-internal v1.14.1/go.mod h1:MaRKkUm5W0goXpeCfT7UZI6fk/L7L7so1lCWt35ZSgc=
+github.com/santhosh-tekuri/jsonschema/v6 v6.0.2 h1:KRzFb2m7YtdldCEkzs6KqmJw4nqEVZGK7IN2kJkjTuQ=
+github.com/santhosh-tekuri/jsonschema/v6 v6.0.2/go.mod h1:JXeL+ps8p7/KNMjDQk3TCwPpBy0wYklyWTfbkIzdIFU=
+github.com/spf13/cast v1.7.1 h1:cuNEagBQEHWN1FnbGEjCXL2szYEXqfJPbP2HNUaca9Y=
+github.com/spf13/cast v1.7.1/go.mod h1:ancEpBxwJDODSW/UG4rDrAqiKolqNNh2DX3mk86cAdo=
+github.com/stretchr/testify v1.11.1 h1:7s2iGBzp5EwR7/aIZr8ao5+dra3wiQyKjjFuvgVKu7U=
+github.com/stretchr/testify v1.11.1/go.mod h1:wZwfW3scLgRK+23gO65QZefKpKQRnfz6sD981Nm4B6U=
+github.com/yosida95/uritemplate/v3 v3.0.2 h1:Ed3Oyj9yrmi9087+NczuL5BwkIc4wvTb5zIM+UJPGz4=
+github.com/yosida95/uritemplate/v3 v3.0.2/go.mod h1:ILOh0sOhIJR3+L/8afwt/kE++YT040gmv5BQTMR2HP4=
+golang.org/x/net v0.33.0 h1:74SYHlV8BIgHIFC/LrYkOGIwL19eTYXQ5wc6TBuO36I=
+golang.org/x/net v0.33.0/go.mod h1:HXLR5J+9DxmrqMwG9qjGCxZ+zKXxBru04zlTvWlWuN4=
+golang.org/x/text v0.21.0 h1:zyQAAkrwaneQ066sspRyJaG9VNi/YJ1NfzcGB3hZ/qo=
+golang.org/x/text v0.21.0/go.mod h1:4IBbMaMmOPCJ8SecivzSH54+73PCFmPWxNTLm+vZkEQ=
+gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
+gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
+resty.dev/v3 v3.0.0-beta.2 h1:xu4mGAdbCLuc3kbk7eddWfWm4JfhwDtdapwss5nCjnQ=
+resty.dev/v3 v3.0.0-beta.2/go.mod h1:OgkqiPvTDtOuV4MGZuUDhwOpkY8enjOsjjMzeOHefy4=
--- a/src/frontend/mcp-server/internal/app.go
+++ b/src/frontend/mcp-server/internal/app.go
@ -0,0 +1,201 @@
+// Package internal provides the scheduler-agent MCP tool handlers.
+package internal
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"strconv"
+
+	"github.com/mark3labs/mcp-go/mcp"
+	sclient "schedulerclient"
+)
+
+// App wires the schedulerclient to the MCP tool handlers.
+type App struct {
+	client sclient.Client
+}
+
+// NewApp creates a new App with a schedulerclient connected to the given base URL.
+func NewApp(baseURL, token string) (*App, error) {
+	hc := sclient.NewClient(baseURL, token)
+	return &App{client: hc}, nil
+}
+
+// ── helpers ─────────────────────────────────────────────────────────────────
+
+// textResult returns an MCP tool result containing the JSON representation of v.
+func textResult(v any) (*mcp.CallToolResult, error) {
+	b, err := json.MarshalIndent(v, "", "  ")
+	if err != nil {
+		return mcp.NewToolResultError(fmt.Sprintf("marshal error: %v", err)), nil
+	}
+	return mcp.NewToolResultText(string(b)), nil
+}
+
+// textResultStr returns an MCP tool result wrapping a plain string.
+func textResultStr(s string) *mcp.CallToolResult {
+	return mcp.NewToolResultText(s)
+}
+
+// getString extracts a string parameter from an MCP tool request.
+func getString(req mcp.CallToolRequest, key string) string {
+	args := req.GetArguments()
+	raw, ok := args[key]
+	if !ok {
+		return ""
+	}
+	s, _ := raw.(string)
+	return s
+}
+
+// getNumber extracts a float64 parameter from an MCP tool request.
+func getNumber(req mcp.CallToolRequest, key string) float64 {
+	args := req.GetArguments()
+	raw, ok := args[key]
+	if !ok {
+		return 0
+	}
+	n, _ := raw.(float64)
+	return n
+}
+
+// ── tool handlers ───────────────────────────────────────────────────────────
+
+func (a *App) HandleListJobs(ctx context.Context, req mcp.CallToolRequest) (*mcp.CallToolResult, error) {
+	filter := sclient.JobFilter{
+		Queue:     getString(req, "queue"),
+		Owner:     getString(req, "owner"),
+		PageToken: getString(req, "page_token"),
+	}
+	if n := getNumber(req, "limit"); n > 0 {
+		filter.Limit = int(n)
+	}
+	if s := getString(req, "status"); s != "" {
+		filter.Statuses = parseStatuses(s)
+	}
+
+	jobs, page, err := a.client.ListJobs(ctx, filter)
+	if err != nil {
+		return mcp.NewToolResultError(fmt.Sprintf("list jobs failed: %v", err)), nil
+	}
+
+	resp := map[string]any{
+		"jobs": jobs,
+		"page": page,
+	}
+	return textResult(resp)
+}
+
+func (a *App) HandleGetJob(ctx context.Context, req mcp.CallToolRequest) (*mcp.CallToolResult, error) {
+	id := getString(req, "id")
+	if id == "" {
+		return mcp.NewToolResultError("job id is required"), nil
+	}
+
+	job, err := a.client.GetJob(ctx, id)
+	if err != nil {
+		return mcp.NewToolResultError(fmt.Sprintf("get job failed: %v", err)), nil
+	}
+	return textResult(job)
+}
+
+func (a *App) HandleListQueues(ctx context.Context, req mcp.CallToolRequest) (*mcp.CallToolResult, error) {
+	queues, err := a.client.ListQueues(ctx)
+	if err != nil {
+		return mcp.NewToolResultError(fmt.Sprintf("list queues failed: %v", err)), nil
+	}
+	return textResult(queues)
+}
+
+func (a *App) HandleListNodes(ctx context.Context, req mcp.CallToolRequest) (*mcp.CallToolResult, error) {
+	nodes, err := a.client.ListNodes(ctx)
+	if err != nil {
+		return mcp.NewToolResultError(fmt.Sprintf("list nodes failed: %v", err)), nil
+	}
+	return textResult(nodes)
+}
+
+func (a *App) HandleJobLogs(ctx context.Context, req mcp.CallToolRequest) (*mcp.CallToolResult, error) {
+	id := getString(req, "id")
+	if id == "" {
+		return mcp.NewToolResultError("job id is required"), nil
+	}
+	opts := sclient.LogOpts{
+		Tail:        int(getNumber(req, "tail")),
+		OffsetBytes: int64(getNumber(req, "offset_bytes")),
+	}
+
+	lines, err := a.client.GetJobLogs(ctx, id, opts)
+	if err != nil {
+		return mcp.NewToolResultError(fmt.Sprintf("get job logs failed: %v", err)), nil
+	}
+	return textResultStr(strconv.Itoa(len(lines)) + " lines\n" + joinLines(lines)), nil
+}
+
+func (a *App) HandleClusterMetrics(ctx context.Context, req mcp.CallToolRequest) (*mcp.CallToolResult, error) {
+	metrics, err := a.client.GetClusterMetrics(ctx)
+	if err != nil {
+		return mcp.NewToolResultError(fmt.Sprintf("get cluster metrics failed: %v", err)), nil
+	}
+	return textResult(metrics)
+}
+
+// ── helpers ─────────────────────────────────────────────────────────────────
+
+// parseStatuses parses a comma-separated list of URM status strings.
+func parseStatuses(s string) []sclient.JobStatus {
+	if s == "" {
+		return nil
+	}
+	parts := splitComma(s)
+	statuses := make([]sclient.JobStatus, 0, len(parts))
+	for _, p := range parts {
+		statuses = append(statuses, sclient.JobStatus(p))
+	}
+	return statuses
+}
+
+// joinLines joins log lines into a single string.
+func joinLines(lines []string) string {
+	if len(lines) == 0 {
+		return ""
+	}
+	if len(lines) == 1 {
+		return lines[0]
+	}
+	// Estimate total length for capacity
+	total := 0
+	for _, l := range lines {
+		total += len(l) + 1
+	}
+	b := make([]byte, 0, total)
+	for i, l := range lines {
+		if i > 0 {
+			b = append(b, '\n')
+		}
+		b = append(b, []byte(l)...)
+	}
+	return string(b)
+}
+
+// splitComma splits a comma-separated string, trimming whitespace.
+func splitComma(s string) []string {
+	if s == "" {
+		return nil
+	}
+	var result []string
+	start := 0
+	for i := 0; i < len(s); i++ {
+		if s[i] == ',' {
+			if i > start {
+				result = append(result, s[start:i])
+			}
+			start = i + 1
+		}
+	}
+	if start < len(s) {
+		result = append(result, s[start:])
+	}
+	return result
+}
--- a/src/frontend/schedulerclient/client.go
+++ b/src/frontend/schedulerclient/client.go
@ -0,0 +1,39 @@
+package schedulerclient
+
+import "context"
+
+// ── Client interface ────────────────────────────────────────────────────────
+
+// Client provides read-only access to the scheduler cluster through a
+// scheduler-agnostic unified resource model (URM).
+//
+// Implementations talk to the apiserver's REST API and authenticate via
+// either a Bearer JWT token or an XCE API Key (xce_ak_ prefix).
+type Client interface {
+	// ── Jobs ──
+
+	// ListJobs returns jobs matching the given filter.
+	// An empty filter returns all visible jobs.
+	ListJobs(ctx context.Context, filter JobFilter) ([]*URMJob, *Page, error)
+
+	// GetJob returns a single job by its ID.
+	GetJob(ctx context.Context, id string) (*URMJob, error)
+
+	// GetJobLogs retrieves output logs for a job.
+	GetJobLogs(ctx context.Context, id string, opts LogOpts) ([]string, error)
+
+	// ── Queues ──
+
+	// ListQueues returns all visible queues with their current state.
+	ListQueues(ctx context.Context) ([]*URMQueue, error)
+
+	// ── Nodes ──
+
+	// ListNodes returns all visible compute nodes with resource usage.
+	ListNodes(ctx context.Context) ([]*URMNode, error)
+
+	// ── Metrics ──
+
+	// GetClusterMetrics returns a snapshot of overall cluster resource usage.
+	GetClusterMetrics(ctx context.Context) (*Metrics, error)
+}
--- a/src/frontend/schedulerclient/go.mod
+++ b/src/frontend/schedulerclient/go.mod
@ -0,0 +1,7 @@
+module schedulerclient
+
+go 1.23.4
+
+require resty.dev/v3 v3.0.0-beta.2
+
+require golang.org/x/net v0.33.0 // indirect
--- a/src/frontend/schedulerclient/go.sum
+++ b/src/frontend/schedulerclient/go.sum
@ -0,0 +1,4 @@
+golang.org/x/net v0.33.0 h1:74SYHlV8BIgHIFC/LrYkOGIwL19eTYXQ5wc6TBuO36I=
+golang.org/x/net v0.33.0/go.mod h1:HXLR5J+9DxmrqMwG9qjGCxZ+zKXxBru04zlTvWlWuN4=
+resty.dev/v3 v3.0.0-beta.2 h1:xu4mGAdbCLuc3kbk7eddWfWm4JfhwDtdapwss5nCjnQ=
+resty.dev/v3 v3.0.0-beta.2/go.mod h1:OgkqiPvTDtOuV4MGZuUDhwOpkY8enjOsjjMzeOHefy4=
--- a/src/frontend/schedulerclient/httpclient.go
+++ b/src/frontend/schedulerclient/httpclient.go
@ -0,0 +1,439 @@
+package schedulerclient
+
+import (
+	"context"
+	"crypto/tls"
+	"encoding/json"
+	"fmt"
+	"math"
+	"net/http"
+	"strconv"
+	"strings"
+	"time"
+
+	"resty.dev/v3"
+)
+
+// ── Client construction ─────────────────────────────────────────────────────
+
+// HTTPClient implements Client by calling the apiserver's REST API.
+// Authenticates via either a JWT Bearer token or an XCE API Key (xce_ak_ prefix).
+type HTTPClient struct {
+	rc      *resty.Client
+	baseURL string
+	token   string
+}
+
+// NewClient creates a new HTTPClient. The token is sent as the Authorization
+// header value exactly as given (no "Bearer " prefix added — the apiserver
+// strips it via its own TrimPrefix logic, so both forms work).
+func NewClient(baseURL, token string) *HTTPClient {
+	c := resty.New()
+	c.SetTimeout(10 * time.Second)
+	c.SetTLSClientConfig(&tls.Config{InsecureSkipVerify: true})
+	return &HTTPClient{
+		rc:      c,
+		baseURL: strings.TrimRight(baseURL, "/"),
+		token:   token,
+	}
+}
+
+// ── Client interface implementation ─────────────────────────────────────────
+
+func (hc *HTTPClient) ListJobs(ctx context.Context, filter JobFilter) ([]*URMJob, *Page, error) {
+	url := hc.baseURL + "/xce/v1/jobs"
+	req := hc.rc.R().SetContext(ctx).SetHeader("Authorization", hc.token)
+
+	if filter.Queue != "" {
+		req.SetQueryParam("queue_name", filter.Queue)
+	}
+	if filter.Owner != "" {
+		req.SetQueryParam("submit_user", filter.Owner)
+	}
+	if filter.Limit > 0 {
+		req.SetQueryParam("limit", strconv.Itoa(filter.Limit))
+	}
+	if filter.PageToken != "" {
+		req.SetQueryParam("page_token", filter.PageToken)
+	}
+	// Note: status filter on URM statuses is applied in post-processing
+	// because the apiserver works with raw status strings, not URM statuses.
+
+	return hc.getJobs(req, url, filter)
+}
+
+func (hc *HTTPClient) GetJob(ctx context.Context, id string) (*URMJob, error) {
+	url := hc.baseURL + "/xce/v1/jobs/" + id
+	req := hc.rc.R().SetContext(ctx).SetHeader("Authorization", hc.token)
+	jobs, _, err := hc.getJobs(req, url, JobFilter{})
+	if err != nil {
+		return nil, err
+	}
+	if len(jobs) == 0 {
+		return nil, fmt.Errorf("job %s not found", id)
+	}
+	return jobs[0], nil
+}
+
+func (hc *HTTPClient) GetJobLogs(ctx context.Context, id string, opts LogOpts) ([]string, error) {
+	url := hc.baseURL + "/xce/v1/jobs/" + id + "/output"
+	req := hc.rc.R().SetContext(ctx).SetHeader("Authorization", hc.token)
+	if opts.Tail > 0 {
+		req.SetQueryParam("tail", strconv.Itoa(opts.Tail))
+	}
+	if opts.OffsetBytes > 0 {
+		req.SetQueryParam("offset", strconv.FormatInt(opts.OffsetBytes, 10))
+	}
+
+	var env apiEnvelope
+	resp, err := req.SetResult(&env).Get(url)
+	if err != nil {
+		return nil, fmt.Errorf("get job logs: %w", err)
+	}
+	if resp.StatusCode() != http.StatusOK {
+		return nil, fmt.Errorf("get job logs: HTTP %d: %s", resp.StatusCode(), env.Msg)
+	}
+	if env.Code != http.StatusOK {
+		return nil, fmt.Errorf("get job logs: api error %d: %s", env.Code, env.Msg)
+	}
+
+	var out struct {
+		Content   string `json:"content"`
+		JobStatus string `json:"job_status"`
+	}
+	if err := json.Unmarshal(env.Data, &out); err != nil {
+		return nil, fmt.Errorf("parse job logs: %w", err)
+	}
+	if out.Content == "" {
+		return []string{}, nil
+	}
+	return strings.Split(strings.TrimSuffix(out.Content, "\n"), "\n"), nil
+}
+
+func (hc *HTTPClient) ListQueues(ctx context.Context) ([]*URMQueue, error) {
+	url := hc.baseURL + "/xce/v1/queues"
+	req := hc.rc.R().SetContext(ctx).SetHeader("Authorization", hc.token)
+
+	var env apiEnvelope
+	resp, err := req.SetResult(&env).Get(url)
+	if err != nil {
+		return nil, fmt.Errorf("list queues: %w", err)
+	}
+	if resp.StatusCode() != http.StatusOK || env.Code != http.StatusOK {
+		return nil, fmt.Errorf("list queues: HTTP %d: %s", resp.StatusCode(), env.Msg)
+	}
+
+	var rows struct {
+		Count int          `json:"count"`
+		Rows  []apiQueue   `json:"rows"`
+	}
+	if err := json.Unmarshal(env.Data, &rows); err != nil {
+		return nil, fmt.Errorf("parse queues: %w", err)
+	}
+
+	queues := make([]*URMQueue, 0, len(rows.Rows))
+	for i := range rows.Rows {
+		q := rows.Rows[i]
+		queues = append(queues, &URMQueue{
+			Name:          q.QueueName,
+			Status:        QueueStatusFromString(q.QueueStatus),
+			RunningJobs:   q.NumRun,
+			PendingJobs:   q.NumPend,
+			SuspendedJobs: q.NumSSusp + q.NumUSusp,
+			NodeCount:     0, // apiserver does not expose queue→node mapping directly
+			Extensions: map[string]any{
+				"raw_status":  q.QueueStatus,
+				"priority":    q.Priority,
+				"num_ssusp":   q.NumSSusp,
+				"num_ususp":   q.NumUSusp,
+				"num_done":    q.NumDone,
+			},
+		})
+	}
+	return queues, nil
+}
+
+func (hc *HTTPClient) ListNodes(ctx context.Context) ([]*URMNode, error) {
+	url := hc.baseURL + "/xce/v1/hosts"
+
+	req := hc.rc.R().SetContext(ctx).SetHeader("Authorization", hc.token)
+
+	var env apiEnvelope
+	resp, err := req.SetResult(&env).Get(url)
+	if err != nil {
+		return nil, fmt.Errorf("list nodes: %w", err)
+	}
+	if resp.StatusCode() != http.StatusOK || env.Code != http.StatusOK {
+		return nil, fmt.Errorf("list nodes: HTTP %d: %s", resp.StatusCode(), env.Msg)
+	}
+
+	var rows struct {
+		Count int        `json:"count"`
+		Rows  []apiHost  `json:"rows"`
+	}
+	if err := json.Unmarshal(env.Data, &rows); err != nil {
+		return nil, fmt.Errorf("parse nodes: %w", err)
+	}
+
+	nodes := make([]*URMNode, 0, len(rows.Rows))
+	for i := range rows.Rows {
+		h := rows.Rows[i]
+		nodes = append(nodes, &URMNode{
+			Name:             h.HostName,
+			Status:           NodeStatusFromString(h.HostStatus),
+			CoresTotal:       h.NumCPUs,
+			CoresUsed:        h.NumRun, // running jobs ≈ cores used
+			MemoryBytesTotal: int64(h.MaxMem) * mbToBytes,
+			MemoryBytesUsed:  0, // apiserver does not expose live used-memory for hosts
+			GPUCount:         0, // TODO: parse from extensions/topology when available
+			Features:         nil,
+			Extensions: map[string]any{
+				"raw_status":  h.HostStatus,
+				"host_type":   h.HostType,
+				"host_model":  h.HostModel,
+				"cpu_factor":  h.CPUFactor,
+				"max_slot":    h.MaxSlot,
+				"max_swap_mb": h.MaxSwap,
+				"max_tmp_mb":  h.MaxTmp,
+				"num_run":     h.NumRun,
+				"num_ssusp":   h.NumSSusp,
+				"num_ususp":   h.NumUSusp,
+			},
+		})
+	}
+	return nodes, nil
+}
+
+func (hc *HTTPClient) GetClusterMetrics(ctx context.Context) (*Metrics, error) {
+	// Fetch hosts to compute total/used resources.
+	nodes, err := hc.ListNodes(ctx)
+	if err != nil {
+		return nil, fmt.Errorf("cluster metrics: %w", err)
+	}
+
+	var (
+		totalCores, usedCores     int
+		totalMem, usedMem        int64
+		totalGPUs                int
+		pendingJobs, runningJobs int
+	)
+
+	for _, n := range nodes {
+		totalCores += n.CoresTotal
+		usedCores += n.CoresUsed
+		totalMem += n.MemoryBytesTotal
+		// usedMem not directly available; try from extensions
+	}
+
+	// Also fetch jobs for running/pending counts — we do this as a cheap
+	// parallel operation to get job counts across all queues.
+	url := hc.baseURL + "/xce/v1/jobs"
+	var env apiEnvelope
+	req := hc.rc.R().SetContext(ctx).SetHeader("Authorization", hc.token)
+	resp, err := req.SetResult(&env).Get(url)
+	if err == nil && resp.StatusCode() == http.StatusOK && env.Code == http.StatusOK {
+		var r struct {
+			Rows []json.RawMessage `json:"rows"`
+		}
+		if json.Unmarshal(env.Data, &r) == nil {
+			for _, raw := range r.Rows {
+				var stub struct {
+					JobStatus string `json:"job_status"`
+				}
+				if json.Unmarshal(raw, &stub) == nil {
+					switch JobStatusFromString(stub.JobStatus) {
+					case JobStatusRunning:
+						runningJobs++
+					case JobStatusPending:
+						pendingJobs++
+					}
+				}
+			}
+		}
+	}
+
+	util := 0.0
+	if totalCores > 0 {
+		util = math.Round(float64(usedCores)/float64(totalCores)*100) / 100
+	}
+
+	return &Metrics{
+		TotalCores:  totalCores,
+		UsedCores:   usedCores,
+		TotalMemory: totalMem,
+		UsedMemory:  usedMem,
+		TotalGPUs:   totalGPUs,
+		UsedGPUs:    0,
+		RunningJobs: runningJobs,
+		PendingJobs: pendingJobs,
+		Utilization: util,
+	}, nil
+}
+
+// ── shared helper ───────────────────────────────────────────────────────────
+
+// getJobs makes an HTTP GET and parses the job list response.
+// When filter.Statuses is non-empty, results are filtered in Go after decoding.
+func (hc *HTTPClient) getJobs(req *resty.Request, url string, filter JobFilter) ([]*URMJob, *Page, error) {
+	var env apiEnvelope
+	resp, err := req.SetResult(&env).Get(url)
+	if err != nil {
+		return nil, nil, fmt.Errorf("list jobs: %w", err)
+	}
+	if resp.StatusCode() != http.StatusOK || env.Code != http.StatusOK {
+		return nil, nil, fmt.Errorf("list jobs: HTTP %d: %s", resp.StatusCode(), env.Msg)
+	}
+
+	var rows struct {
+		Count     int       `json:"count"`
+		Rows      []apiJob  `json:"rows"`
+		PageToken string    `json:"page_token"`
+	}
+	if err := json.Unmarshal(env.Data, &rows); err != nil {
+		return nil, nil, fmt.Errorf("parse jobs: %w", err)
+	}
+
+	jobs := make([]*URMJob, 0, len(rows.Rows))
+	for i := range rows.Rows {
+		j := rows.Rows[i]
+		urmStatus := JobStatusFromString(j.JobStatus)
+
+		// Apply URM status filter in post-processing
+		if len(filter.Statuses) > 0 {
+			matched := false
+			for _, s := range filter.Statuses {
+				if urmStatus == s {
+					matched = true
+					break
+				}
+			}
+			if !matched {
+				continue
+			}
+		}
+
+		// Parse exit code
+		var exitCode *int
+		if j.ExitStatus != "" {
+			if code, err := strconv.Atoi(j.ExitStatus); err == nil {
+				exitCode = &code
+			}
+		}
+
+		// Parse timestamps
+		var submitTime time.Time
+		if j.SubmitTime > 0 {
+			submitTime = time.Unix(j.SubmitTime, 0)
+		}
+		var startTime *time.Time
+		if j.StartTime > 0 {
+			t := time.Unix(j.StartTime, 0)
+			startTime = &t
+		}
+
+		// Walltime ≈ total CPU time available.  For XCE jobs, run_limit
+		// is available as a string ("HH:MM") and MaxNumProcessors gives
+		// slot count.  We store what we can.
+		walltime := 0 // TODO: parse run_limit string when available
+
+		// Memory: MaxMem from apiserver is in MB (standard HPC convention).
+		memBytes := int64(math.Round(j.MaxMem * mbToBytes))
+
+		urmJob := &URMJob{
+			ID:              strconv.FormatInt(j.JobID, 10),
+			Name:            j.JobName,
+			Status:          urmStatus,
+			RawStatus:       j.JobStatus,
+			Queue:           j.QueueName,
+			Nodes:           j.NumHosts,
+			Cores:           j.NumProcessors,
+			GPUs:            0, // TODO: parse from resreq/extensions
+			MemoryBytes:     memBytes,
+			WalltimeSeconds: walltime,
+			SubmissionTime:  submitTime,
+			StartTime:       startTime,
+			ExitCode:        exitCode,
+			Owner:           j.SubmitUser,
+			Extensions: map[string]any{
+				"command":       j.Command,
+				"project_name":  j.ProjectName,
+				"pending_reason": j.PendingReason,
+				"resreq":        j.Resreq,
+				"exit_status_raw": j.ExitStatus,
+				"num_processors": j.NumProcessors,
+				"submit_time":   j.SubmitTime,
+				"start_time":    j.StartTime,
+				"end_time":      j.EndTime,
+			},
+		}
+
+		jobs = append(jobs, urmJob)
+	}
+
+	page := &Page{
+		Total:         rows.Count,
+		NextPageToken: rows.PageToken,
+	}
+	return jobs, page, nil
+}
+
+// ── internal types for JSON decoding (apiserver response shapes) ────────────
+
+const mbToBytes = 1_000_000 // MaxMem from apiserver is in MB → stored as bytes in URM
+
+// apiEnvelope is the standard top-level response: {"code":200,"msg":"ok","data":...}
+type apiEnvelope struct {
+	Code int             `json:"code"`
+	Msg  string          `json:"msg"`
+	Data json.RawMessage `json:"data"`
+}
+
+// apiJob mirrors the fields of apiserver's QueryJob that we need for URM conversion.
+type apiJob struct {
+	JobID          int64   `json:"job_id"`
+	SubmitUser     string  `json:"submit_user"`
+	JobStatus      string  `json:"job_status"`
+	JobName        string  `json:"job_name"`
+	Command        string  `json:"command"`
+	QueueName      string  `json:"queue_name"`
+	ProjectName    string  `json:"project_name"`
+	NumProcessors  int     `json:"num_processors"`
+	NumHosts       int     `json:"num_hosts"`
+	SubmitTime     int64   `json:"submit_time"`
+	StartTime      int64   `json:"start_time"`
+	EndTime        int64   `json:"end_time"`
+	MaxMem         float64 `json:"max_mem"`
+	ExitStatus     string  `json:"exit_status"`
+	PendingReason  string  `json:"pending_reason"`
+	Resreq         string  `json:"resreq"`
+}
+
+// apiQueue mirrors fields of DBQueueInfo needed for URM conversion.
+type apiQueue struct {
+	QueueName   string `json:"queue_name"`
+	QueueStatus string `json:"queue_status"`
+	Priority    int    `json:"priority"`
+	NumPend     int    `json:"num_pend"`
+	NumRun      int    `json:"num_run"`
+	NumSSusp    int    `json:"num_ssusp"`
+	NumUSusp    int    `json:"num_ususp"`
+	NumDone     int    `json:"num_done"`
+}
+
+// apiHost mirrors fields of XCEHost needed for URM conversion.
+type apiHost struct {
+	HostName   string  `json:"host_name"`
+	HostStatus string  `json:"host_status"`
+	HostType   string  `json:"host_type"`
+	HostModel  string  `json:"host_model"`
+	CPUFactor  float64 `json:"cpu_factor"`
+	NumCPUs    int     `json:"num_cpus"`
+	MaxSlot    int     `json:"max_slot"`
+	MaxMem     int     `json:"max_mem"`
+	MaxSwap    int     `json:"max_swap"`
+	MaxTmp     int     `json:"max_tmp"`
+	NumRun     int     `json:"num_run"`
+	NumSSusp   int     `json:"num_ssusp"`
+	NumUSusp   int     `json:"num_ususp"`
+	NumDone    int     `json:"num_done"`
+}
--- a/src/frontend/schedulerclient/statusmap.go
+++ b/src/frontend/schedulerclient/statusmap.go
@ -0,0 +1,93 @@
+package schedulerclient
+
+import "strings"
+
+// ── JobStatus mapping ───────────────────────────────────────────────────────
+
+// JobStatusFromString maps the apiserver's job status string to a URM JobStatus.
+//
+// The input is the output of apiserver models.JobStateStr() — one of:
+//
+//	PEND, PSUSP, RUN, USUSP, SSUSP, RUNKWN, EXIT, DONE, UNKWN, ERR, TRANSIENT
+func JobStatusFromString(s string) JobStatus {
+	switch s {
+	case "PEND", "PSUSP":
+		return JobStatusPending
+	case "RUN", "RUNKWN":
+		return JobStatusRunning
+	case "USUSP", "SSUSP":
+		return JobStatusSuspended
+	case "EXIT":
+		return JobStatusFailed
+	case "DONE":
+		return JobStatusSucceeded
+	default:
+		// UNKWN, ERR, TRANSIENT, empty, or unknown values → unknown
+		return JobStatusUnknown
+	}
+}
+
+// ── QueueStatus mapping ─────────────────────────────────────────────────────
+
+// QueueStatusFromString maps the apiserver's composite queue status string
+// (e.g. "open:active", "closed:inactive_adm") to a URM status:
+//
+//	active   — queue is open and accepting jobs
+//	inactive — queue is administratively or window-closed
+//	draining — open but has an inactive-time window active (will close)
+//	unknown  — unrecognised format
+func QueueStatusFromString(s string) string {
+	if s == "" {
+		return "unknown"
+	}
+
+	// Format is "<open|closed>:<active|inactive_adm|inactive_win>"
+	parts := strings.SplitN(s, ":", 2)
+	if len(parts) < 1 {
+		return "unknown"
+	}
+
+	openClosed := parts[0]
+	var inactiveKind string
+	if len(parts) > 1 {
+		inactiveKind = parts[1]
+	}
+
+	switch openClosed {
+	case "open":
+		if inactiveKind == "inactive_adm" || inactiveKind == "inactive_win" {
+			return "draining"
+		}
+		return "active"
+	case "closed":
+		return "inactive"
+	default:
+		return "unknown"
+	}
+}
+
+// ── NodeStatus mapping ─────────────────────────────────────────────────────
+
+// NodeStatusFromString maps the apiserver's host status string (output of
+// models.HostStateStr()) to a URM node status:
+//
+//	idle    — OK, accepting work
+//	busy    — running jobs
+//	down    — CLOSED, CLOSED_BY_ADMIN
+//	drain   — LOCKU, LOCKW (not accepting new work, existing jobs finish)
+//	unknown — UNAVAIL or unrecognised
+func NodeStatusFromString(s string) string {
+	switch s {
+	case "OK":
+		return "idle"
+	case "BUSY":
+		return "busy"
+	case "CLOSED", "CLOSED_BY_ADMIN":
+		return "down"
+	case "LOCKU", "LOCKW":
+		return "drain"
+	default:
+		// UNAVAIL, empty, or unknown → unknown
+		return "unknown"
+	}
+}
--- a/src/frontend/schedulerclient/statusmap_test.go
+++ b/src/frontend/schedulerclient/statusmap_test.go
@ -0,0 +1,94 @@
+package schedulerclient
+
+import "testing"
+
+// ── JobStatus mapping tests ─────────────────────────────────────────────────
+
+func TestJobStatusFromString(t *testing.T) {
+	tests := []struct {
+		input string
+		want  JobStatus
+	}{
+		// PEND → pending
+		{"PEND", JobStatusPending},
+		{"PSUSP", JobStatusPending}, // user-suspended before start → still pending
+		// RUN → running
+		{"RUN", JobStatusRunning},
+		{"RUNKWN", JobStatusRunning}, // running but state uncertain → treat as running
+		// suspended
+		{"USUSP", JobStatusSuspended}, // user-suspended while running
+		{"SSUSP", JobStatusSuspended}, // system-suspended while running
+		// terminal
+		{"EXIT", JobStatusFailed},
+		{"DONE", JobStatusSucceeded},
+		// unknown / error
+		{"UNKWN", JobStatusUnknown},
+		{"ERR", JobStatusUnknown},
+		// edge cases
+		{"TRANSIENT", JobStatusUnknown},
+		{"", JobStatusUnknown},
+		{"NONEXISTENT", JobStatusUnknown},
+	}
+	for _, tt := range tests {
+		t.Run(tt.input, func(t *testing.T) {
+			got := JobStatusFromString(tt.input)
+			if got != tt.want {
+				t.Errorf("JobStatusFromString(%q) = %q, want %q", tt.input, got, tt.want)
+			}
+		})
+	}
+}
+
+// ── QueueStatus mapping tests ────────────────────────────────────────────────
+
+func TestQueueStatusFromString(t *testing.T) {
+	tests := []struct {
+		input string
+		want  string
+	}{
+		{"open:active", "active"},
+		{"closed:active", "inactive"},
+		{"open:inactive_adm", "draining"},
+		{"open:inactive_win", "draining"},
+		{"closed:inactive_adm", "inactive"},
+		{"closed:inactive_win", "inactive"},
+		{"open", "active"},
+		{"closed", "inactive"},
+		{"", "unknown"},
+	}
+	for _, tt := range tests {
+		t.Run(tt.input, func(t *testing.T) {
+			got := QueueStatusFromString(tt.input)
+			if got != tt.want {
+				t.Errorf("QueueStatusFromString(%q) = %q, want %q", tt.input, got, tt.want)
+			}
+		})
+	}
+}
+
+// ── NodeStatus mapping tests ────────────────────────────────────────────────
+
+func TestNodeStatusFromString(t *testing.T) {
+	tests := []struct {
+		input string
+		want  string
+	}{
+		{"OK", "idle"},
+		{"BUSY", "busy"},
+		{"CLOSED", "down"},
+		{"CLOSED_BY_ADMIN", "down"},
+		{"LOCKU", "drain"},
+		{"LOCKW", "drain"},
+		{"UNAVAIL", "unknown"},
+		{"", "unknown"},
+		{"NONEXISTENT", "unknown"},
+	}
+	for _, tt := range tests {
+		t.Run(tt.input, func(t *testing.T) {
+			got := NodeStatusFromString(tt.input)
+			if got != tt.want {
+				t.Errorf("NodeStatusFromString(%q) = %q, want %q", tt.input, got, tt.want)
+			}
+		})
+	}
+}
--- a/src/frontend/schedulerclient/types.go
+++ b/src/frontend/schedulerclient/types.go
@ -0,0 +1,104 @@
+// Package schedulerclient provides a read-only Go client for the XCE scheduler API,
+// exposing a unified resource model (URM) that maps the cluster's native concepts
+// (jobs, queues, nodes) into a scheduler-agnostic representation.
+//
+// This is the library that the Phase 0 MCP Server will use to talk to the apiserver.
+package schedulerclient
+
+import "time"
+
+// ── Unified Resource Model (URM) types ──────────────────────────────────────
+
+// JobStatus is the canonical status of a job across all scheduler backends.
+type JobStatus string
+
+const (
+	JobStatusPending   JobStatus = "pending"
+	JobStatusRunning   JobStatus = "running"
+	JobStatusSucceeded JobStatus = "succeeded"
+	JobStatusFailed    JobStatus = "failed"
+	JobStatusCancelled JobStatus = "cancelled"
+	JobStatusSuspended JobStatus = "suspended"
+	JobStatusUnknown   JobStatus = "unknown"
+)
+
+// URMJob is the unified representation of a scheduler job.
+type URMJob struct {
+	ID              string         `json:"id"`
+	Name            string         `json:"name"`
+	Status          JobStatus      `json:"status"`
+	RawStatus       string         `json:"raw_status"` // backend-native status string
+	Queue           string         `json:"queue"`
+	Nodes           int            `json:"nodes"`
+	Cores           int            `json:"cores"`
+	GPUs            int            `json:"gpus"`
+	MemoryBytes     int64          `json:"memory_bytes"`
+	WalltimeSeconds int            `json:"walltime_seconds"`
+	SubmissionTime  time.Time      `json:"submission_time"`
+	StartTime       *time.Time     `json:"start_time,omitempty"`
+	ExitCode        *int           `json:"exit_code,omitempty"`
+	Owner           string         `json:"owner"`
+	LogsURL         string         `json:"logs_url,omitempty"`
+	Extensions      map[string]any `json:"extensions,omitempty"`
+}
+
+// URMQueue is the unified representation of a scheduler queue/partition.
+type URMQueue struct {
+	Name               string         `json:"name"`
+	Status             string         `json:"status"` // active | inactive | draining
+	RunningJobs        int            `json:"running_jobs"`
+	PendingJobs        int            `json:"pending_jobs"`
+	SuspendedJobs      int            `json:"suspended_jobs"`
+	NodeCount          int            `json:"node_count"`
+	MaxWalltimeSeconds int            `json:"max_walltime_seconds"`
+	Extensions         map[string]any `json:"extensions,omitempty"`
+}
+
+// URMNode is the unified representation of a compute node.
+type URMNode struct {
+	Name             string         `json:"name"`
+	Status           string         `json:"status"` // idle | busy | down | drain | unknown
+	CoresTotal       int            `json:"cores_total"`
+	CoresUsed        int            `json:"cores_used"`
+	MemoryBytesTotal int64          `json:"memory_bytes_total"`
+	MemoryBytesUsed  int64          `json:"memory_bytes_used"`
+	GPUCount         int            `json:"gpu_count"`
+	Features         []string       `json:"features"`
+	Extensions       map[string]any `json:"extensions,omitempty"`
+}
+
+// ── Filter / input types ────────────────────────────────────────────────────
+
+// JobFilter specifies criteria for listing jobs.
+type JobFilter struct {
+	Statuses  []JobStatus // filter by URM status(es)
+	Queue     string      // filter by queue name
+	Owner     string      // filter by owner username
+	Limit     int         // max results (0 = default)
+	PageToken string      // pagination cursor from a prior response
+}
+
+// LogOpts specifies how to retrieve job logs.
+type LogOpts struct {
+	Tail        int   // last N lines (0 = all)
+	OffsetBytes int64 // byte offset to start from
+}
+
+// Page carries pagination metadata returned with list operations.
+type Page struct {
+	Total         int    `json:"total"`
+	NextPageToken string `json:"next_page_token,omitempty"`
+}
+
+// Metrics is a snapshot of overall cluster resource usage.
+type Metrics struct {
+	TotalCores  int     `json:"total_cores"`
+	UsedCores   int     `json:"used_cores"`
+	TotalMemory int64   `json:"total_memory_bytes"`
+	UsedMemory  int64   `json:"used_memory_bytes"`
+	TotalGPUs   int     `json:"total_gpus"`
+	UsedGPUs    int     `json:"used_gpus"`
+	RunningJobs int     `json:"running_jobs"`
+	PendingJobs int     `json:"pending_jobs"`
+	Utilization float64 `json:"utilization"` // 0.0 – 1.0
+}