manus-mcp-server
Enables Claude to generate images by wrapping the Manus API through a remote MCP server with OAuth authentication.
README
manus-mcp-server
A Remote MCP server that wraps the Manus API and exposes a generate_image
tool, so Claude (web/app) can request images and have Manus generate them.
How it works
Manus's API is a general task API, not a dedicated image endpoint. This server:
- Receives a
generate_imagetool call from Claude (prompt, optional style/aspect ratio). - Creates a Manus task asking the agent to produce an image and return it as a file.
- Polls the task until it completes (or times out after 4 minutes).
- Returns the resulting file URL(s) back to Claude as the tool result.
1. Deploy on Railway
- Push this repo to GitHub.
- In Railway: New Project → Deploy from GitHub repo → select this repo.
- Railway will detect
railway.jsonand use Nixpacks to runnpm install && npm run build, thennpm start. - In the Railway project, go to Variables and set:
MANUS_API_KEY— your Manus API key.MANUS_API_BASE_URL— defaults tohttps://api.manus.im/v1; check your Manus API docs/dashboard and adjust if different.ADMIN_PASSWORD— a strong password only you know. This gates the/authorizelogin page so a stranger who finds your server's URL can't mint themselves an access token.PUBLIC_BASE_URL— set this to your Railway public domain once you have it (step 5 below), e.g.https://manus-mcp-server-production.up.railway.app. If left unset, the server tries to infer it from request headers, which usually works on Railway but is safer to set explicitly.PORT— Railway sets this automatically; no action needed.
- Once deployed, go to Settings → Networking → Generate Domain to get a public HTTPS URL, e.g.
https://manus-mcp-server-production.up.railway.app. - Confirm it's alive: visit
https://<your-domain>/health— should return{"status":"ok"}.
2. Verify the Manus API endpoint shape
The exact paths (/tasks, response field names like task_id vs id) can
vary by account/API version. Before relying on this in production:
- Check
https://manus.im/docs/integrations/manus-api(or your account's API reference) for the current create-task and get-task endpoints and response fields. - If they differ from what's in
src/manusClient.ts, adjust the constants at the top of that file (CREATE_TASK_PATH,GET_TASK_PATH, and the field names read increateImageTask/getTask) and redeploy. - Easiest way to check without guessing: make one manual
curlcall to create a task and print the raw JSON, then match the field names.
3. Connect it to Claude
This server now implements a real OAuth 2.0 flow (dynamic client registration + authorization code + PKCE), which is what Claude's "Add custom connector" screen expects.
In Claude (web or app): Settings → Connectors → Add custom connector.
- URL:
https://<your-railway-domain>/mcp - Claude will auto-discover the OAuth endpoints via
https://<your-railway-domain>/.well-known/oauth-authorization-server, register itself as a client, then redirect you to a login page. - On that login page, enter the
ADMIN_PASSWORDyou set in Railway. This approves the connection and Claude receives an access token.
Once connected, you can just ask Claude for an image and it will call the
generate_image tool automatically.
Notes on this OAuth implementation
- It's intentionally minimal: in-memory storage, single admin password, tokens expire after 1 hour and refresh automatically via the refresh token Claude stores.
- Restarting/redeploying the Railway service clears all registered clients and tokens — you'll need to remove and re-add the connector in Claude if that happens.
- This is appropriate for personal, single-user use. It is not a general-purpose identity provider — don't hand this URL to other people.
Local development
npm install
cp .env.example .env # fill in your real values
npm run dev
Server runs at http://localhost:3000/mcp.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。