@tonyclaw/agent-inspector 2.0.33 → 2.0.35

package/src/mcp/server.ts

@@ -1,8 +1,7 @@
 /**
  * MCP (Model Context Protocol) server for agent-inspector.
  *
- * Exposes `inspector_*` tools that delegate to existing `/api/*` endpoints
- * via internal HTTP loopback (see `loopback.ts` for the rationale). Mounted
+ * Exposes `inspector_*` tools and resources. Mounted
  * at `POST /api/mcp` in the existing TanStack Start process using HTTP Streamable
  * transport. localhost-only, no auth in v1 — see
  * `openspec/changes/add-mcp-server/design.md` decisions D1, D2, D9.
@@ -11,8 +10,8 @@
  * instantiated once per process on first request and reused for all
  * subsequent requests (stateless mode — no `sessionIdGenerator`).
  *
- * Tool implementations live in `./toolHandlers.ts` so they can be unit
- * tested with a mock `callApi`. This file just wires them to the SDK.
+ * Tool implementations live in `./toolHandlers.ts`; this file wires them to
+ * the SDK and exposes MCP resources/prompts.
  */
 
 import { z } from "zod";
@@ -30,18 +29,24 @@ import {
   AddInspectorGroupSessionInputSchema,
   DeleteInspectorGroupOptionsSchema,
   GroupEvidenceExportOptionsSchema,
-  InspectorGroupSchema,
   InspectorGroupStatusSchema,
-  InspectorGroupsListResponseSchema,
   UpdateInspectorGroupInputSchema,
 } from "../lib/groupContract";
-import {
-  InspectorRunSchema,
-  InspectorRunStatusSchema,
-  UpdateInspectorRunInputSchema,
-} from "../lib/runContract";
+import { InspectorRunStatusSchema, UpdateInspectorRunInputSchema } from "../lib/runContract";
 import { setCurrentPort } from "../lib/serverPort";
 import { JsonValueSchema } from "../proxy/schemas";
+import {
+  getInspectorGroup,
+  listInspectorGroups,
+  readInspectorGroupEvidence,
+  readInspectorGroupSessions,
+} from "../services/groups";
+import {
+  getInspectorRun,
+  listInspectorRecentFailures,
+  listInspectorRuns,
+  readInspectorRunEvidence,
+} from "../services/runs";
 import { callApi } from "./loopback";
 import {
   addGroupSessionImpl,
@@ -179,22 +184,10 @@ Either preview field is null when the body is unknown format, unparseable, or th
 const PROVIDER_WRITE_WARNING =
   "⚠ This tool mutates provider configuration. Confirm with the user before invoking.";
 
-const RunsListResponseSchema = z.object({
-  runs: z.array(InspectorRunSchema),
-});
-
 const LogResourceListResponseSchema = z.object({
   logs: z.array(z.object({ id: z.number().int().positive() })),
 });
 
-const EvidenceReadResponseSchema = z.object({
-  markdown: z.string(),
-});
-
-const GroupEvidenceReadResponseSchema = z.object({
-  markdown: z.string(),
-});
-
 type ResourceDescriptor = {
   uri: string;
   name: string;
@@ -236,6 +229,14 @@ async function readApiJsonResource(uri: URL, path: string): Promise<ReadResource
   }
 }
 
+function readServiceJsonResource<T>(
+  uri: URL,
+  result: { ok: true; value: T } | { ok: false; message: string },
+): ReadResourceResult {
+  if (!result.ok) return textResource(uri, "text/plain", result.message);
+  return jsonResource(uri, result.value);
+}
+
 async function listSessionResources(): Promise<ListResourcesResult> {
   try {
     const response = await callApi("/api/sessions");
@@ -274,105 +275,65 @@ async function listLogResources(): Promise<ListResourcesResult> {
   }
 }
 
-async function listRunResources(): Promise<ListResourcesResult> {
-  try {
-    const response = await callApi("/api/runs");
-    if (!response.ok) return listResult([]);
-    const parsed = RunsListResponseSchema.safeParse(await response.json());
-    if (!parsed.success) return listResult([]);
-    return listResult(
-      parsed.data.runs.map((run) => ({
-        uri: `inspector://runs/${encodeURIComponent(run.id)}`,
-        name: `run:${run.id}`,
-        title: run.title,
-        mimeType: "application/json",
-      })),
-    );
-  } catch {
-    return listResult([]);
-  }
+function listRunResources(): ListResourcesResult {
+  return listResult(
+    listInspectorRuns().runs.map((run) => ({
+      uri: `inspector://runs/${encodeURIComponent(run.id)}`,
+      name: `run:${run.id}`,
+      title: run.title,
+      mimeType: "application/json",
+    })),
+  );
 }
 
-async function listGroupResources(): Promise<ListResourcesResult> {
-  try {
-    const response = await callApi("/api/groups");
-    if (!response.ok) return listResult([]);
-    const parsed = InspectorGroupsListResponseSchema.safeParse(await response.json());
-    if (!parsed.success) return listResult([]);
-    return listResult(
-      parsed.data.groups.map((group) => ({
-        uri: `inspector://groups/${encodeURIComponent(group.id)}`,
-        name: `group:${group.id}`,
-        title: group.title,
-        mimeType: "application/json",
-      })),
-    );
-  } catch {
-    return listResult([]);
-  }
+function listGroupResources(): ListResourcesResult {
+  return listResult(
+    listInspectorGroups().groups.map((group) => ({
+      uri: `inspector://groups/${encodeURIComponent(group.id)}`,
+      name: `group:${group.id}`,
+      title: group.title,
+      mimeType: "application/json",
+    })),
+  );
 }
 
-async function listGroupSessionResources(): Promise<ListResourcesResult> {
-  try {
-    const response = await callApi("/api/groups");
-    if (!response.ok) return listResult([]);
-    const parsed = InspectorGroupsListResponseSchema.safeParse(await response.json());
-    if (!parsed.success) return listResult([]);
-    return listResult(
-      parsed.data.groups.map((group) => ({
-        uri: `inspector://groups/${encodeURIComponent(group.id)}/sessions`,
-        name: `group-sessions:${group.id}`,
-        title: `${group.title} sessions`,
-        mimeType: "application/json",
-      })),
-    );
-  } catch {
-    return listResult([]);
-  }
+function listGroupSessionResources(): ListResourcesResult {
+  return listResult(
+    listInspectorGroups().groups.map((group) => ({
+      uri: `inspector://groups/${encodeURIComponent(group.id)}/sessions`,
+      name: `group-sessions:${group.id}`,
+      title: `${group.title} sessions`,
+      mimeType: "application/json",
+    })),
+  );
 }
 
-async function listGroupEvidenceResources(): Promise<ListResourcesResult> {
-  try {
-    const response = await callApi("/api/groups");
-    if (!response.ok) return listResult([]);
-    const parsed = InspectorGroupsListResponseSchema.safeParse(await response.json());
-    if (!parsed.success) return listResult([]);
-    return listResult(
-      parsed.data.groups
-        .filter((group) => group.evidence !== null)
-        .map((group) => ({
-          uri: `inspector://groups/${encodeURIComponent(group.id)}/evidence`,
-          name: `group-evidence:${group.id}`,
-          title: `${group.title} evidence pack`,
-          mimeType: "text/markdown",
-          description: group.evidence?.markdownPath,
-        })),
-    );
-  } catch {
-    return listResult([]);
-  }
+function listGroupEvidenceResources(): ListResourcesResult {
+  return listResult(
+    listInspectorGroups()
+      .groups.filter((group) => group.evidence !== null)
+      .map((group) => ({
+        uri: `inspector://groups/${encodeURIComponent(group.id)}/evidence`,
+        name: `group-evidence:${group.id}`,
+        title: `${group.title} evidence pack`,
+        mimeType: "text/markdown",
+        description: group.evidence?.markdownPath,
+      })),
+  );
 }
 
-async function listEvidenceResources(): Promise<ListResourcesResult> {
-  try {
-    const response = await callApi("/api/runs");
-    if (!response.ok) return listResult([]);
-    const parsed = RunsListResponseSchema.safeParse(await response.json());
-    if (!parsed.success) return listResult([]);
-    return listResult(
-      parsed.data.runs
-        .filter((run) => run.evidence !== null)
-        .map((run) => ({
-          uri: `inspector://evidence/${encodeURIComponent(run.id)}`,
-          name: `evidence:${run.id}`,
-          title: `${run.title} evidence pack`,
-          mimeType: "text/markdown",
-          description: run.evidence?.markdownPath,
-        })),
-    );
-  } catch {
-    return listResult([]);
-  }
+function listEvidenceResources(): ListResourcesResult {
+  return listResult(
+    listInspectorRuns()
+      .runs.filter((run) => run.evidence !== null)
+      .map((run) => ({
+        uri: `inspector://evidence/${encodeURIComponent(run.id)}`,
+        name: `evidence:${run.id}`,
+        title: `${run.title} evidence pack`,
+        mimeType: "text/markdown",
+        description: run.evidence?.markdownPath,
+      })),
+  );
 }
 
 function userPrompt(text: string): GetPromptResult {
@@ -531,7 +492,7 @@ function registerTools(server: McpServer): void {
         metadata: z.record(z.string(), JsonValueSchema).optional().describe("JSON metadata."),
       }),
     },
-    (args) => safeCall(() => createRunImpl(callApi, args)),
+    (args) => safeCall(() => createRunImpl(args)),
   );
 
   server.registerTool(
@@ -544,7 +505,7 @@ function registerTools(server: McpServer): void {
         runId: z.string().min(1).describe("The Inspector run id."),
       }),
     },
-    (args) => safeCall(() => getRunImpl(callApi, args)),
+    (args) => safeCall(() => getRunImpl(args)),
   );
 
   server.registerTool(
@@ -557,7 +518,7 @@ function registerTools(server: McpServer): void {
         runId: z.string().min(1).describe("The Inspector run id."),
       }),
     },
-    (args) => safeCall(() => updateRunImpl(callApi, args)),
+    (args) => safeCall(() => updateRunImpl(args)),
   );
 
   server.registerTool(
@@ -581,7 +542,7 @@ function registerTools(server: McpServer): void {
           .describe("How many compact log summaries to include in the evidence pack. Max 50."),
       }),
     },
-    (args) => safeCall(() => exportEvidenceImpl(callApi, args)),
+    (args) => safeCall(() => exportEvidenceImpl(args)),
   );
 
   server.registerTool(
@@ -600,7 +561,7 @@ function registerTools(server: McpServer): void {
           .describe("Maximum failures to return. Default 5, max 20."),
       }),
     },
-    (args) => safeCall(() => getRecentFailuresImpl(callApi, args)),
+    (args) => safeCall(() => getRecentFailuresImpl(args)),
   );
 
   server.registerTool(
@@ -611,7 +572,7 @@ function registerTools(server: McpServer): void {
         "Lists declared Inspector groups. Groups aggregate multiple sessions/runs, commonly one evaluation matrix spanning several providers or models.",
       inputSchema: z.object({}),
     },
-    () => safeCall(() => listGroupsImpl(callApi)),
+    () => safeCall(() => listGroupsImpl()),
   );
 
   server.registerTool(
@@ -634,7 +595,7 @@ function registerTools(server: McpServer): void {
         metadata: z.record(z.string(), JsonValueSchema).optional().describe("JSON metadata."),
       }),
     },
-    (args) => safeCall(() => createGroupImpl(callApi, args)),
+    (args) => safeCall(() => createGroupImpl(args)),
   );
 
   server.registerTool(
@@ -647,7 +608,7 @@ function registerTools(server: McpServer): void {
         groupId: z.string().min(1).describe("The Inspector group id."),
       }),
     },
-    (args) => safeCall(() => getGroupImpl(callApi, args)),
+    (args) => safeCall(() => getGroupImpl(args)),
   );
 
   server.registerTool(
@@ -660,7 +621,7 @@ function registerTools(server: McpServer): void {
         groupId: z.string().min(1).describe("The Inspector group id."),
       }),
     },
-    (args) => safeCall(() => updateGroupImpl(callApi, args)),
+    (args) => safeCall(() => updateGroupImpl(args)),
   );
 
   server.registerTool(
@@ -673,7 +634,7 @@ function registerTools(server: McpServer): void {
         groupId: z.string().min(1).describe("The Inspector group id."),
       }),
     },
-    (args) => safeCall(() => deleteGroupImpl(callApi, args)),
+    (args) => safeCall(() => deleteGroupImpl(args)),
   );
 
   server.registerTool(
@@ -686,7 +647,7 @@ function registerTools(server: McpServer): void {
         groupId: z.string().min(1).describe("The Inspector group id."),
       }),
     },
-    (args) => safeCall(() => addGroupSessionImpl(callApi, args)),
+    (args) => safeCall(() => addGroupSessionImpl(args)),
   );
 
   server.registerTool(
@@ -699,7 +660,7 @@ function registerTools(server: McpServer): void {
         groupId: z.string().min(1).describe("The Inspector group id."),
       }),
     },
-    (args) => safeCall(() => exportGroupEvidenceImpl(callApi, args)),
+    (args) => safeCall(() => exportGroupEvidenceImpl(args)),
   );
 
   server.registerTool(
@@ -942,7 +903,7 @@ function registerResources(server: McpServer): void {
       description: "Declared Inspector evaluation runs.",
       mimeType: "application/json",
     },
-    (uri) => readApiJsonResource(uri, "/api/runs"),
+    (uri) => jsonResource(uri, listInspectorRuns()),
   );
 
   server.registerResource(
@@ -953,7 +914,7 @@ function registerResources(server: McpServer): void {
       description: "Declared Inspector groups aggregating multiple sessions or runs.",
       mimeType: "application/json",
     },
-    (uri) => readApiJsonResource(uri, "/api/groups"),
+    (uri) => jsonResource(uri, listInspectorGroups()),
   );
 
   server.registerResource(
@@ -967,7 +928,7 @@ function registerResources(server: McpServer): void {
     (uri, variables) => {
       const groupId = variableString(variables, "groupId");
       if (groupId === null) return textResource(uri, "text/plain", "Missing groupId");
-      return readApiJsonResource(uri, `/api/groups/${encodeURIComponent(groupId)}`);
+      return readServiceJsonResource(uri, getInspectorGroup(groupId));
     },
   );
 
@@ -984,7 +945,7 @@ function registerResources(server: McpServer): void {
     (uri, variables) => {
       const groupId = variableString(variables, "groupId");
       if (groupId === null) return textResource(uri, "text/plain", "Missing groupId");
-      return readApiJsonResource(uri, `/api/groups/${encodeURIComponent(groupId)}/sessions`);
+      return readServiceJsonResource(uri, readInspectorGroupSessions(groupId));
     },
   );
 
@@ -997,7 +958,7 @@ function registerResources(server: McpServer): void {
         "Recent declared runs with failed/cancelled status or non-success exported evidence.",
       mimeType: "application/json",
     },
-    (uri) => readApiJsonResource(uri, "/api/runs?failures=1&limit=10"),
+    (uri) => jsonResource(uri, listInspectorRecentFailures(10)),
   );
 
   server.registerResource(
@@ -1011,7 +972,7 @@ function registerResources(server: McpServer): void {
     (uri, variables) => {
       const runId = variableString(variables, "runId");
       if (runId === null) return textResource(uri, "text/plain", "Missing runId");
-      return readApiJsonResource(uri, `/api/runs/${encodeURIComponent(runId)}`);
+      return readServiceJsonResource(uri, getInspectorRun(runId));
     },
   );
 
@@ -1024,24 +985,12 @@ function registerResources(server: McpServer): void {
         "Markdown evidence pack for a declared Inspector run. Generate it first with inspector_export_evidence.",
       mimeType: "text/markdown",
     },
-    async (uri, variables) => {
+    (uri, variables) => {
       const runId = variableString(variables, "runId");
       if (runId === null) return textResource(uri, "text/plain", "Missing runId");
-      const path = `/api/runs/${encodeURIComponent(runId)}/evidence`;
-      try {
-        const response = await callApi(path);
-        if (!response.ok) {
-          return textResource(uri, "text/plain", `GET ${path} returned ${response.status}`);
-        }
-        const parsed = EvidenceReadResponseSchema.safeParse(await response.json());
-        if (!parsed.success) {
-          return textResource(uri, "text/plain", "Evidence endpoint returned an unparseable body");
-        }
-        return textResource(uri, "text/markdown", parsed.data.markdown);
-      } catch (err) {
-        const message = err instanceof Error ? err.message : String(err);
-        return textResource(uri, "text/plain", `Resource read failed: ${message}`);
-      }
+      const evidence = readInspectorRunEvidence(runId);
+      if (!evidence.ok) return textResource(uri, "text/plain", evidence.message);
+      return textResource(uri, "text/markdown", evidence.value.markdown);
     },
   );
 
@@ -1056,24 +1005,12 @@ function registerResources(server: McpServer): void {
         "Markdown evidence pack for an Inspector group. Generate it first with inspector_export_group_evidence.",
       mimeType: "text/markdown",
     },
-    async (uri, variables) => {
+    (uri, variables) => {
       const groupId = variableString(variables, "groupId");
       if (groupId === null) return textResource(uri, "text/plain", "Missing groupId");
-      const path = `/api/groups/${encodeURIComponent(groupId)}/evidence`;
-      try {
-        const response = await callApi(path);
-        if (!response.ok) {
-          return textResource(uri, "text/plain", `GET ${path} returned ${response.status}`);
-        }
-        const parsed = GroupEvidenceReadResponseSchema.safeParse(await response.json());
-        if (!parsed.success) {
-          return textResource(uri, "text/plain", "Evidence endpoint returned an unparseable body");
-        }
-        return textResource(uri, "text/markdown", parsed.data.markdown);
-      } catch (err) {
-        const message = err instanceof Error ? err.message : String(err);
-        return textResource(uri, "text/plain", `Resource read failed: ${message}`);
-      }
+      const evidence = readInspectorGroupEvidence(groupId);
+      if (!evidence.ok) return textResource(uri, "text/plain", evidence.message);
+      return textResource(uri, "text/markdown", evidence.value.markdown);
     },
   );
 }