fix(agent): preserve cloud MCP relays

Generated-By: PostHog Code
Task-Id: c6d78416-85c8-4304-bd69-e309562468c4

Author: richardsolomou

docs/cloud-mcp-relay.md

@@ -22,7 +22,9 @@ The built-in `posthog-code-computer-use` designation resolves to the packaged
 PostHog local-tools MCP server instead of `~/.claude.json`. It is added to new
 cloud runs when the desktop **Computer use** setting is enabled and exposes
 only the macOS computer tools. The same relay approval policy applies, so tool
-calls execute only while the creating desktop remains connected.
+calls execute only while the creating desktop remains connected. Runs with
+imported or relayed MCP servers skip pre-warmed sandbox reuse because the
+server list is not known when the warm run starts.
 
 ## Problem
 

packages/agent/src/adapters/codex-app-server/local-tools-mcp.test.ts

@@ -3,11 +3,17 @@ import { LOCAL_TOOLS_MCP_NAME } from "../local-tools";
 import { buildComputerUseStdioServer } from "../local-tools/stdio-server";
 import { buildLocalToolsServer } from "./local-tools-mcp";
 
-// The dist asset isn't on the walk-up path in unit tests, so make existsSync
-// succeed; nothing spawns the script — we only inspect the path.
+// Isolate bundled-script resolution and the sandbox env file; the test only
+// inspects the generated stdio config.
 vi.mock("node:fs", async (importActual) => {
   const actual = await importActual<typeof import("node:fs")>();
-  return { ...actual, existsSync: vi.fn().mockReturnValue(true) };
+  return {
+    ...actual,
+    existsSync: vi.fn().mockReturnValue(true),
+    readFileSync: vi.fn(() => {
+      throw new Error("sandbox env file unavailable");
+    }),
+  };
 });
 
 describe("buildLocalToolsServer", () => {

packages/core/src/task-detail/taskCreationSaga.test.ts

@@ -1,4 +1,5 @@
 import type { SessionService } from "@posthog/core/sessions/sessionService";
+import { CLOUD_COMPUTER_USE_MCP_NAME } from "@posthog/shared";
 import type { Task, TaskRun } from "@posthog/shared/domain-types";
 import { beforeEach, describe, expect, it, vi } from "vitest";
 import type {
@@ -45,6 +46,7 @@ const sessionService = {
   disconnectFromTask: vi.fn(),
   rememberInitialCloudPrompt: vi.fn(),
   markTaskCreationInFlight: vi.fn(),
+  designateRelayedMcpServers: vi.fn().mockResolvedValue(undefined),
 } as unknown as SessionService;
 
 const createTask = (overrides: Partial<Task> = {}): Task => ({
@@ -733,6 +735,35 @@ describe("TaskCreationSaga", () => {
     },
   );
 
+  it("uses a cold run when desktop-relayed MCP servers are requested", async () => {
+    const createdTask = createTask();
+    const startedTask = createTask({ latest_run: createRun() });
+    const createTaskMock = vi.fn().mockResolvedValue(createdTask);
+    const createTaskRunMock = vi.fn().mockResolvedValue(createRun());
+    const saga = makeSaga({
+      createTask: createTaskMock,
+      createTaskRun: createTaskRunMock,
+      startTaskRun: vi.fn().mockResolvedValue(startedTask),
+    });
+    const relayedMcpServers = [{ name: CLOUD_COMPUTER_USE_MCP_NAME }];
+
+    const result = await saga.run({
+      content: "Ship the fix",
+      repository: "posthog/posthog",
+      workspaceMode: "cloud",
+      branch: "main",
+      relayedMcpServers,
+    });
+
+    expect(result.success).toBe(true);
+    expect(mockHost.takeWarmTaskLease).not.toHaveBeenCalled();
+    expect(createTaskMock.mock.calls[0][0].branch).toBeUndefined();
+    expect(createTaskRunMock).toHaveBeenCalledWith(
+      "task-123",
+      expect.objectContaining({ relayedMcpServers }),
+    );
+  });
+
   it("reuses a warm run built from the selected custom image", async () => {
     mockHost.takeWarmTaskLease.mockReturnValue({
       taskId: "warm-task",

packages/core/src/task-detail/taskCreationSaga.ts

@@ -685,6 +685,10 @@ export class TaskCreationSaga extends Saga<
       augmented,
     };
 
+    if (input.importedMcpServers?.length || input.relayedMcpServers?.length) {
+      return { ...base, suppressWarmReuse: true };
+    }
+
     const lease = input.repository
       ? this.deps.host.takeWarmTaskLease({
           repository: input.repository,