OpenAPI Slice
@vvarp
An MCP (Model Context Protocol) server that helps you work with large OpenAPI specifications by extracting only the relevant portions for specific endpoints. This is particularly useful when working with LLMs that have context limitations - instead of loading an entire large Open
Overview
What is OpenAPI Slice MCP?
OpenAPI Slice MCP is a Model Context Protocol server that extracts only the relevant portions of large OpenAPI specifications for specific endpoints. It helps developers and LLMs work with large specs by reducing context size through automatic dependency resolution, supporting both local files and remote URLs.
How to use OpenAPI Slice MCP?
Run the server with uvx openapi-slice-mcp. It communicates via STDIO and provides tools to load specs (from file or URL), list endpoints, extract endpoint slices, and check server status.
Key features of OpenAPI Slice MCP
- Endpoint-specific extraction with automatic dependency resolution
- Supports local YAML/JSON files and remote URLs (HTTP/HTTPS)
- Output in YAML or JSON format
- Discovery tool to list all endpoints in a loaded spec
- Recursively includes all referenced components (schemas, parameters, etc.)
Use cases of OpenAPI Slice MCP
- Extracting a minimal spec for a single API endpoint to fit within LLM context limits
- Reducing token usage when sending OpenAPI definitions to AI assistants
- Isolating endpoint definitions for testing or documentation generation
FAQ from OpenAPI Slice MCP
What makes OpenAPI Slice MCP different from loading the full spec?
It extracts only the requested endpoint and its dependencies, unlike loading the entire OpenAPI specification. This is especially useful when working with LLMs that have context limitations.
What are the runtime requirements?
The server is run via uvx (from the uv package manager) and requires Python. No additional dependencies are listed beyond those bundled in the package.
Where does the data live?
Specifications can be loaded from local files (YAML/JSON) or fetched from remote URLs. No data is persisted on the server beyond the currently loaded spec.
What transport and authentication does it use?
The server uses STDIO transport and does not mention any authentication mechanism. It is designed for local or trusted setups.
Can I get the output in different formats?
Yes, the extract_endpoint_slice tool accepts an output_format parameter that can be set to "yaml" (default) or "json".
