Files
shumai/packages/e2e/workflow/mcp.test.ts
T
Yiling-J 851ae49c84 feat(mcp): support tool-level directTools array and 3-state UI configuration (#297)
* feat(mcp): support tool-level directTools array and 3-state UI configuration

* refactor(mcp): refine tool hint explanation and separate server instructions block

* style(mcp): display tool count in header title and split option hints into 3 lines

* style(mcp): update tools description paragraph and place bulk buttons on a separate row

* fix(mcp): enforce excludeTools in proxy path and align direct-tool counts

- Filter the per-agent tool registry by excludeTools at build time (DB cache
  and lazy-connect paths), so excluded tools are unreachable through the mcp
  proxy (call, search, list, describe) — matching pi-mcp-adapter where
  isToolAllowed gates both direct and gateway exposure.
- Add shared isDirectTool predicate (matches original or prefixed name) used
  by both buildDirectTools and the proxy description counts, so the
  advertised direct/proxy counts can no longer drift from actual
  registration; toolCount now reflects allowed tools only.
- Remove the legacy boolean directTools branch (no legacy data exists).
- Drop the now-unused mcp_excluded_tools_hint i18n key and regenerate
  Paraglide messages.
- Extend unit tests: excluded tools are not counted, listed, searched, or
  callable via the proxy; prefixed-name directTools configs count correctly.
2026-08-11 15:43:27 +08:00

335 lines
11 KiB
TypeScript

import { describe, it, expect, beforeAll, afterAll, vi } from 'vitest'
import { prisma } from '@shumai/db'
import { setupTestDbHooks } from '@shumai/db/test'
import { workflowService, TaskQueueAgent } from '@shumai/workflow-core'
import { initAgentWorkflows } from '@shumai/agent'
import { s3Service } from '@shumai/core/src/s3/s3'
import { mcpService } from '@shumai/core/src/mcp/mcp-service'
import {
startTestMcpServer,
standardTestTools,
type RunningTestMcpServer,
} from '@shumai/core/src/mcp/mcp-test-server'
import { AgentHarness } from '@earendil-works/pi-agent-core'
import { fileURLToPath } from 'url'
import * as path from 'path'
const currentDir = path.dirname(fileURLToPath(import.meta.url))
const agentWorkflowsPath = path.resolve(currentDir, '../../../apps/agent/src/workflows.ts')
describe.each(['local', 'temporal'] as const)('Workflow E2E - MCP tools (executor: %s)', (mode) => {
setupTestDbHooks()
let agentWorkerPromise: Promise<void> | null = null
let mcpTestServer: RunningTestMcpServer
beforeAll(async () => {
process.env.S3_BUCKET = 'shumai-e2e-test-bucket-mcp'
process.env.GEMINI_API_KEY = 'dummy-key'
// Real in-process MCP server — only AI calls are mocked. It self-reports
// the name 'e2e-mcp' so discovery auto-fills the DB name/tool prefixes
// exactly as the tests expect.
mcpTestServer = await startTestMcpServer({
tools: standardTestTools(),
serverInfo: { name: 'e2e-mcp' },
})
workflowService.setExecutorType(mode)
initAgentWorkflows()
// Mock ONLY the AI model call. The mocked prompt locates the real `mcp`
// proxy tool on the harness and executes it against the real MCP server,
// so the full createAgentSession → buildAgentTools → proxy execute →
// callMcpTool → MCP server path is exercised.
vi.spyOn(AgentHarness.prototype, 'prompt').mockImplementation(async function (
this: AgentHarness,
) {
const tools = this.getTools()
const mcpTool = tools.find((t) => t.name === 'mcp')
if (!mcpTool) {
return {
content: [{ type: 'text', text: 'NO_MCP_TOOL' }],
usage: { input: 10, output: 20 },
stopReason: 'stop',
} as unknown as Awaited<ReturnType<typeof AgentHarness.prototype.prompt>>
}
const result = await (
mcpTool.execute as unknown as (
id: string,
params: { server: string; tool: string; args: Record<string, unknown> },
) => Promise<{ content: Array<{ type: string; text?: string }> }>
)('e2e-call', {
server: 'e2e-mcp',
tool: 'echo',
args: { text: 'hello from mcp e2e' },
})
const text = result.content
.map((c) => ('text' in c && typeof c.text === 'string' ? c.text : ''))
.join('\n')
return {
content: [{ type: 'text', text }],
usage: { input: 10, output: 20 },
stopReason: 'stop',
} as unknown as Awaited<ReturnType<typeof AgentHarness.prototype.prompt>>
})
if (mode === 'temporal') {
console.log('Starting background worker for MCP Temporal E2E tests...')
agentWorkerPromise = workflowService.startWorkers(TaskQueueAgent, {
workflowsPath: agentWorkflowsPath,
})
await new Promise((resolve) => setTimeout(resolve, 2000))
} else {
console.log('Starting local workflow service polling...')
workflowService.start()
}
})
afterAll(async () => {
if (mode === 'temporal') {
console.log('Shutting down Temporal workers...')
await workflowService.shutdownWorkers()
await Promise.all([agentWorkerPromise].filter(Boolean))
}
workflowService.close()
vi.restoreAllMocks()
await mcpService.closeAllConnections()
await mcpTestServer.stop()
try {
console.log('Cleaning up local E2E storage files...')
await s3Service.deletePrefix('shumai-e2e-test-bucket-mcp', '')
} catch (err) {
console.error('Failed to clean up E2E storage folder:', err)
}
})
async function seedChatSetup() {
const team = await prisma.team.create({
data: { name: `MCP E2E Team (${mode})` },
})
const project = await prisma.project.create({
data: { name: 'MCP E2E Project', teamId: team.id },
})
const agentUser = await prisma.user.create({
data: {
name: 'MCP E2E Agent',
email: `mcp-e2e-${mode}@shumai.ai`,
type: 'agent',
},
})
await prisma.teamMember.create({
data: { teamId: team.id, userId: agentUser.id, role: 'editor' },
})
const provider = await prisma.provider.create({
data: {
name: 'google',
teamId: team.id,
config: { api: 'google-generative-ai', apiKey: 'dummy-google-api-key' },
},
})
const model = await prisma.model.create({
data: {
modelId: 'gemini',
name: 'Gemini',
providerId: provider.id,
config: {
input: ['text'],
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
contextWindow: 8192,
maxTokens: 2048,
reasoning: false,
},
},
})
await prisma.agent.create({
data: {
id: agentUser.id,
teamId: team.id,
type: 'chat',
enabled: true,
providerId: provider.id,
modelId: model.id,
config: { provider: 'google', model: 'gemini' },
},
})
const asset = await prisma.asset.create({
data: {
name: 'mcp-chat-asset.png',
type: 'file',
status: 'uploaded',
mediaType: 'image/png',
projectId: project.id,
},
})
const regularUser = await prisma.user.create({
data: {
name: 'MCP Regular User',
email: `mcp-user-${mode}@example.com`,
type: 'human',
},
})
await prisma.teamMember.create({
data: { teamId: team.id, userId: regularUser.id, role: 'reviewer' },
})
await prisma.agentSession.create({
data: {
id: `mcp-session-${mode}`,
agentId: agentUser.id,
userId: regularUser.id,
cwd: process.cwd(),
assetId: asset.id,
},
})
return { team, project, agentUser, asset, regularUser }
}
async function createMcpServerAndAssign(teamId: string, agentId: string, config?: object) {
const server = await prisma.mcpServer.create({
data: {
name: 'e2e-mcp',
url: mcpTestServer.url,
teamId,
config: (config ?? {}) as PrismaJson.McpServerConfig,
},
})
await prisma.agentMcpServer.create({ data: { agentId, mcpServerId: server.id } })
// Mirrors the admin flow: refresh the server so its tools are cached in
// DB and warmed in the in-process registry (buildAgentTools needs the
// cache to register direct tools before any chat turn runs).
await mcpService.refreshServer(server.id)
return server
}
it('runs a chat that calls the mcp proxy tool against a real MCP server', async () => {
const { team, project, agentUser, asset, regularUser } = await seedChatSetup()
await createMcpServerAndAssign(team.id, agentUser.id)
const userComment = await prisma.assetComment.create({
data: {
assetId: asset.id,
message: 'Please use the MCP server',
creatorId: regularUser.id,
sessionId: `mcp-session-${mode}`,
},
})
const task = await prisma.workflowTask.create({
data: {
type: 'chat',
status: 'pending',
assetId: asset.id,
projectId: project.id,
teamId: team.id,
payload: {
projectId: project.id,
agent: {
agentId: agentUser.id,
sessionId: `mcp-session-${mode}`,
userCommentId: userComment.id,
userId: regularUser.id,
},
},
},
})
console.log(`Submitted MCP E2E Chat Task. ID: ${task.id}. Awaiting completion...`)
const completedTask = await workflowService.executeWait(task, 45000)
expect(completedTask.status).toBe('completed')
const comments = await prisma.assetComment.findMany({
where: { assetId: asset.id, replyToId: userComment.id },
})
expect(comments.length).toBeGreaterThan(0)
// The reply is the mocked model's text, which is the REAL MCP echo result.
expect(comments[0].message).toContain('echo:hello from mcp e2e')
}, 50000)
it('runs a chat with direct-tools mode (server-prefixed native tools)', async () => {
const { team, project, agentUser, asset, regularUser } = await seedChatSetup()
await createMcpServerAndAssign(team.id, agentUser.id, { directTools: ['echo'] })
// For direct mode the model would call `e2e_mcp_echo` directly; simulate
// that by executing the direct tool on the harness.
const directSpy = vi
.spyOn(AgentHarness.prototype, 'prompt')
.mockImplementationOnce(async function (this: AgentHarness) {
const tools = this.getTools()
const echoTool = tools.find((t) => t.name === 'e2e_mcp_echo')
if (!echoTool) {
return {
content: [{ type: 'text', text: 'NO_DIRECT_TOOL' }],
usage: { input: 1, output: 1 },
stopReason: 'stop',
} as unknown as Awaited<ReturnType<typeof AgentHarness.prototype.prompt>>
}
const result = await (
echoTool.execute as unknown as (
id: string,
params: Record<string, unknown>,
) => Promise<{ content: Array<{ type: string; text?: string }> }>
)('e2e-direct', {
text: 'direct call',
})
const text = result.content
.map((c) => ('text' in c && typeof c.text === 'string' ? c.text : ''))
.join('\n')
return {
content: [{ type: 'text', text }],
usage: { input: 1, output: 1 },
stopReason: 'stop',
} as unknown as Awaited<ReturnType<typeof AgentHarness.prototype.prompt>>
})
const userComment = await prisma.assetComment.create({
data: {
assetId: asset.id,
message: 'Use the direct tool',
creatorId: regularUser.id,
sessionId: `mcp-session-${mode}`,
},
})
const task = await prisma.workflowTask.create({
data: {
type: 'chat',
status: 'pending',
assetId: asset.id,
projectId: project.id,
teamId: team.id,
payload: {
projectId: project.id,
agent: {
agentId: agentUser.id,
sessionId: `mcp-session-${mode}`,
userCommentId: userComment.id,
userId: regularUser.id,
},
},
},
})
const completedTask = await workflowService.executeWait(task, 45000)
expect(completedTask.status).toBe('completed')
const comments = await prisma.assetComment.findMany({
where: { assetId: asset.id, replyToId: userComment.id },
})
expect(comments.length).toBeGreaterThan(0)
expect(comments[0].message).toContain('echo:direct call')
directSpy.mockRestore()
}, 50000)
})