auto-mcp-server
Enables LLMs to interact with any REST API that has an OpenAPI specification by providing a lightweight MCP server that translates between natural language and API calls.
README
auto-mcp-server
You've got a web app with a REST API. Maybe it's a Django app and you've generated an OpenAPI spec using drf-spectacular.
Now you want to let an LLM talk to it. You need an MCP server to handle the translation between the two.
graph LR
A[🤖 LLM] <--> B[🔌 MCP Server] <--> C[🌐 API Server]
auto-mcp-server is a lightweight MCP server that makes your Web App accessible by LLMs. It was originally made for https://syracuse.1145.am (code: https://github.com/alanbuxton/syracuse-neo) but could be used for any server that has an OpenAPI spec. Not sure how good it would be for heavy production use, but it's certainly good enough for getting started.
It comes in two flavours:
- A local MCP server that uses STDIO and can work with e.g. the free tier of Claude Desktop
- A remote MCP server that uses Streamable HTTP and can be deployed on any web host to your liking. It's not really doing any streaming, but that's ok if you're just dealing with individual REST API calls.
Using this repo
- Clone it
- Ensure you have
uvinstalled (see https://docs.astral.sh/uv/getting-started/installation/) - Copy
.env.sampleto.envand configure according to your settings - Test with MCP Inspector
npx @modelcontextprotocol/inspector(see https://modelcontextprotocol.io/legacy/tools/inspector)
Local MCP Server (STDIO)
With a local MCP server you don't have to start it, you just have to configure your client app (e.g. Claude Desktop) to know where to find the code. Here is a sample claude_desktop_config.json
{
"mcpServers": {
"Your app name": {
"command": "/path/to/uv",
"args": [
"--directory",
"/path/to/auto-mcp-server",
"run",
"stdio_mcp_server.py"
]
}
}
}
Connection info for MCP Inspector:
- Transport Type: STDIO
- Command:
/path/to/uv - Arguments:
--directory /path/to/auto-mcp-server run stdio_mcp_server.py
Remote MCP Server (Stateless Streaming HTTP)
This needs to be started before MCP Inspector can connect to it. Start it with:
uv run stateless_streaming_http_mcp_server.py
Connection info for MCP Inspector:
- Transport Type: Streamable HTTP
- URL: host:port/mcp, e.g.
http://127.0.0.1:9000/mcp - Anyone making calls that require authorization will need to add their API key in the Authentication section
The Stateless Streaming MCP Server also serves an mcp.json at .well-known/mcp.json (in this example it would be http://127.0.0.1/.well-known/mcp.json )
Distinguishing tools that need auth
Each tool is tagged with whether it requires authentication, derived from the OpenAPI security rules (an operation's security overrides the spec-level security, and an explicit empty security: [] means no auth). This is exposed two ways:
- In
tools/list, each tool carries_meta: { "requiresAuth": true | false }. - In
.well-known/mcp.json, the same_metaflag is present, and thesecurityblock is only attached to tools that actually require auth.
Clients can use this to call public endpoints without prompting the user for a token, and only request credentials for tools that need them.
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。