I'm thrilled to share my latest open-source project that bridges the gap between REST APIs and Large Language Models!
Swagger MCP Adapter is a TypeScript-based MCP (Model Context Protocol) server that seamlessly integrates Swagger/OpenAPI specifications with LLMs like Claude, enabling them to interact with your APIs through standardized tools.
🔥 Key Features:
🔹 4 Powerful MCP Tools:
-
list_services- Clean markdown-formatted API service listing -
get_service_information- Detailed info for specific endpoints -
get_all_service_information- Comprehensive API documentation -
get_cache_information- Cache monitoring & performance metrics
🔹 Production-Ready:
- Intelligent caching system with TTL-based expiration
- Optimized ESM builds for production deployment
- Full Claude Desktop integration
- Comprehensive error handling & logging
🔹 Developer Experience:
- Hot reloading during development
- TypeScript-first with full type safety
- Comprehensive test suite (6/6 tests passing)
- Clean, maintainable architecture
💡 Real-World Impact:
This tool transforms how developers and LLMs interact with APIs:
- For Developers: Instant API exploration and documentation
- For LLMs: Structured, validated API interactions
- For Teams: Faster API integration and testing
🛠️ Tech Stack:
TypeScript, Node.js, MCP SDK, Swagger Parser, Zod, Pino, tsup, Vitest
📈 Perfect for:
- API documentation teams
- LLM application developers
- DevOps engineers
- Backend developers
- AI/ML engineers
🔗 Links:
- GitHub: https://github.com/serifcolakel/swagger-mcp-adapter
- Demo: Try with Claude Desktop, GitHub Copilot or any LLM!
- Demo Video on YouTube: https://www.youtube.com/watch?v=Alo34U5axbI
- Docs: Comprehensive README with setup guide
What do you think? Have you worked with MCP servers or API integration tools? I'd love to hear your thoughts! 💬
Top comments (2)
How does the adapter handle authentication defined in the OpenAPI spec (e.g., OAuth2, API keys) when an LLM invokes protected endpoints?
Hi, thank you for your feedback. The MCP Provider does not call your service. Instead, the MCP provides your service implementation to the LLM (MCP Client). This way, the LLM knows how to use it and can apply it to your codebase.