ChatATP Studio MCP Server
Exposes ChatATP Studio's backend API as MCP tools over Streamable HTTP, enabling clients to manage agents, knowledge bases, MCP connections, HTTP APIs, LLM providers, messaging platforms, and teams with forwarded authorization.
README
ChatATP Studio MCP Server
Remote FastMCP server that turns the Studio Django /dapi/ API into MCP tools.
Public URL this project is built for:
https://mcp.chat-atp.com/studio/mcp
Clients (Cursor, Claude, Copilot, ChatATP agents) connect there over Streamable HTTP. Every tool call is proxied to your existing backend with the caller's Authorization header.
What you get
Action-oriented tools over:
- Agents (
list_agents,create_agent,preview_agent, ...) - Knowledge bases + agent attachments
- MCP server definitions + connections
- HTTP API tools +
execute_http_api_connection - LLM providers + API-key configs
- Messaging platforms
- Teams / whoami
Deletes and disconnects require confirm=true.
Local run
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
export STUDIO_API_URL=https://chatatp-agent-builder-backend.onrender.com
export STUDIO_TOKEN=chatatp_sk_... # optional fallback
uvicorn app:app --host 0.0.0.0 --port 8000
Check:
curl http://localhost:8000/health
MCP endpoint:
http://localhost:8000/studio/mcp
Auth
- Preferred: the MCP client sends
Authorization: Bearer <studio jwt or chatatp_sk_...>
The server forwards that header to/dapi/. - Fallback:
STUDIO_TOKEN/STUDIO_API_KEYin the environment (single-tenant / internal).
Do not bake customer tokens into the image.
Deploy at mcp.chat-atp.com/studio/mcp
Point the mcp.chat-atp.com host at this service with the path intact.
MCP_PATH defaults to /studio/mcp, so the process should own the hostname (or receive /studio/mcp unstripped).
Docker
docker build -t chatatp-studio-mcp .
docker run --rm -p 8000:8000 \
-e STUDIO_API_URL=https://chatatp-agent-builder-backend.onrender.com \
-e STUDIO_TOKEN=... \
chatatp-studio-mcp
Render
This repo includes render.yaml. After deploy, put Cloudflare / DNS:
mcp.chat-atp.com → <your-render-service>.onrender.com
If you already have something else on mcp.chat-atp.com and only want /studio/* here, reverse-proxy without stripping the prefix:
location /studio/ {
proxy_pass http://studio-mcp:8000;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header Authorization $http_authorization;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
proxy_buffering off;
proxy_read_timeout 120s;
}
location /health {
proxy_pass http://studio-mcp:8000/health;
}
Caddy:
mcp.chat-atp.com {
reverse_proxy /studio/* localhost:8000
reverse_proxy /health localhost:8000
}
If your proxy strips /studio, set MCP_PATH=/mcp so the public URL is still /studio/mcp after the strip. Do not do both.
Connect a client
Cursor / Claude Desktop / any Streamable HTTP MCP client:
{
"mcpServers": {
"chatatp-studio": {
"url": "https://mcp.chat-atp.com/studio/mcp",
"headers": {
"Authorization": "Bearer chatatp_sk_..."
}
}
}
}
Inspector:
npx @modelcontextprotocol/inspector
# then connect to http://localhost:8000/studio/mcp
Env
| Variable | Default | Purpose |
|---|---|---|
STUDIO_API_URL |
Render backend URL | Django base URL |
STUDIO_TOKEN |
empty | Fallback auth |
MCP_PATH |
/studio/mcp |
Public MCP path |
PORT |
8000 |
Listen port |
STUDIO_TIMEOUT |
30 |
Upstream timeout seconds |
MCP_STATELESS |
true |
Stateless Streamable HTTP (good behind load balancers) |
Notes
- Paths marked inferred in the Studio CLI (
/dapi/mcp/servers/,/dapi/http-api/tools/,/dapi/llm/configs/, agent preview) live instudio_mcp/endpoints.py. If a route 404s, change only that file. - Document upload is not exposed yet (multipart). Add a dedicated tool when you want file ingest from MCP.
- Copilot assistant endpoints (
/dapi/assistant/...) are intentionally not wrapped so this server does not recurse into Copilot.
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。