Neura Market
OpenAPI AnyApi logo

OpenAPI AnyApi

Free

使用内置语义搜索与大型 OpenAPI 文档交互，可自定义前缀。

FreeFree tier
Type
Open Source
Visit Website GitHub (40 stars)

About OpenAPI AnyApi

MCP Server: Scalable OpenAPI Endpoint Discovery and API Request Tool. This MCP server enables Claude (or any MCP client) to discover and call any API endpoint through semantic search. It intelligently chunks OpenAPI specifications to handle large API documentation (100KB+) and provides built-in request execution capabilities. The server uses in-memory FAISS vector search with a MiniLM-L3 embedding model to find relevant endpoints by natural language queries. It is designed for integrating private APIs with Claude Desktop, supporting remote OpenAPI JSON files as the source (no local file system access needed). Configuration is done via environment variables (OPENAPI_JSON_DOCS_URL, MCP_API_PREFIX, GLOBAL_TOOL_PROMPT). Note: There is a cold start penalty (~15s for model loading) if not using the pre-loaded Docker image, and the Docker image size is large (~3.76GB with models). Not supported on linux/arm/v7.

Key Features

Uses remote OpenAPI JSON file as source – no local file system access, no updating required for API changes
Semantic search using optimized MiniLM-L3 embedding model (43MB) with in-memory FAISS vector search for instant endpoint discovery
Endpoint-based chunking of OpenAPI specs – handles 100KB+ documents without loss of endpoint context
FastAPI-based server with async support
Customizable tool namespace via MCP_API_PREFIX environment variable
GLOBAL_TOOL_PROMPT to prepend context to all tool descriptions – helps Claude select the right tool accurately
Built-in request execution – constructs and executes actual RESTful API calls
Supports any MCP client (e.g., Claude Desktop)

Pros & Cons

Pros
  • Handles very large OpenAPI documents (100KB+) that cause errors in standard MCP
  • Fast in-memory semantic search returns relevant endpoints in milliseconds
  • No local file storage needed – reads from remote JSON, updates automatically
  • Asynchronous FastAPI server for concurrent requests
  • Easy to customize tool names and prompts for better AI selection
  • Open source and free
Cons
  • Cold start penalty of ~15 seconds for model loading if not using pre-downloaded Docker image
  • Docker image size is large (~3.76GB with pre-downloaded models, ~2GB without)
  • Does not support linux/arm/v7 due to build issues with Transformer library
  • Dependency on Hugging Face for model download (if not using pre-built Docker image)

Best For

Integrating private APIs with Claude DesktopDiscovering and calling API endpoints via natural language queriesHandling large OpenAPI specifications that exceed MCP client limitsExposing multiple API endpoints from a single OpenAPI spec to an AI assistant

FAQ

How does this MCP server work?
The server takes a remote OpenAPI JSON specification, splits it into endpoint-level chunks, and indexes them with in-memory FAISS using MiniLM-L3 embeddings. When an MCP client (like Claude Desktop) sends a natural language query, it performs semantic search to find the most relevant endpoints, returns their complete documentation, and then allows the client to execute the actual RESTful API request through a separate tool.
What environment variables are available?
OPENAPI_JSON_DOCS_URL (default: https://api.staging.readymojo.com/openapi.json), MCP_API_PREFIX (default 'any_openapi', customizes tool namespace), and GLOBAL_TOOL_PROMPT (optional text prepended to all tool descriptions to help Claude select the right tool).
Can I use this with my private API?
Yes. The server is designed for private APIs. You provide the URL to your OpenAPI JSON spec, and it will index the endpoints. The spec is fetched remotely, so you don't need local file access.
What are the limitations?
Cold start penalty (~15s model loading), no linux/arm/v7 support, large Docker image size (~3.76GB with models), and dependency on Hugging Face for model downloads (unless using pre-loaded Docker image).