OpenAPI to MCP Server

简体中文 | English

This project spins up a Model Context Protocol (MCP) server directly from the supplied openapi.yaml, using uv for dependency management. Every HTTP endpoint becomes an MCP tool that AnythingLLM (or any MCP-aware client) can call.

Quick Start

1. Install dependencies

   uv sync

2. Configure

   base_url: "https://localhost:12001" verify_ssl: false openapi_path: "../openapi.yaml" # Optional, defaults to the repo root copy log_file: "../logs/openapi2mcpserver.log" # Optional, enables file logging log_level: "INFO" # Optional, supports DEBUG/INFO/WARNING/ERROR auth: mode: "basic" # Options: none/basic/bearer/header/query username: "your-username" password: "your-password" # basic_token: "cHJlLWVuY29kZWQtY3JlZHM=" # extra_headers: # X-API-Version: "1" # default_headers: # X-Request-Source: "mcp"

3. Start the MCP server

   uv run openapi2mcpserver

Using This MCP Server in AnythingLLM

The steps below target the AnythingLLM desktop plugin system and list the default paths for macOS and Windows. Create the file if it does not exist and replace the sample paths with the values on your machine.

macOS

1. Open ~/Library/Application Support/anythingllm-desktop/storage/plugins/anythingllm_mcp_servers.json

2. Add or merge the following configuration:

   { "mcpServers": { "rhino-agent": { "command": "/Users/<你的用户名>/.local/bin/uv", "args": ["run", "--project", "/Users/<你的用户名>/workspace/AIProjects/openapi2mcpserver", "openapi2mcpserver"] } } }

Windows

1. Open %APPDATA%\anythingllm-desktop\storage\plugins\anythingllm_mcp_servers.json

2. If you use uv.exe, replace the path with the output of where uv, for example:

   { "mcpServers": { "rhino-agent": { "command": "C:\\Users\\<YourUser>\\AppData\\Roaming\\Python\\Scripts\\uv.exe", "args": ["run", "--project", "C:\\Users\\<YourUser>\\workspace\\AIProjects\\openapi2mcpserver", "openapi2mcpserver"] } } }

Restart AnythingLLM (or reload plugins in settings) after saving, and you should see rhino-agent available in the tool list for your workspace.

Highlights

• Automatically parses every OpenAPI path and registers a matching MCP tool

• Exposes OpenAPI_GetSchema to inspect components.schemas

• Tool inputs cover path params, query params, headers, and request bodies

• Flexible authentication (none/basic/bearer/custom header/query) plus global default headers

• Structured logging to both console and file for easier LLM troubleshooting

Layout

Runtime Notes

At startup the server ingests the OpenAPI description, builds JSON Schema input for every tool, and forwards calls to the upstream API. Responses (status, headers, and JSON/text bodies) are returned verbatim to the MCP client so you can chain follow-up actions such as polling task status.