Hi HN,
I built MCP Discovery API, an agent-native discovery layer for Model Context Protocol servers.
The problem: AI agents can only use tools you pre-configure. They can't discover new capabilities dynamically.
The solution: A semantic search API that returns the best MCP server for any task.
Example:
Query: "I need to send emails with templates"
Response: Gmail MCP server (87% confidence) + install: npx -y @anthropic/mcp-server-gmail
Features:
- Semantic search using OpenAI embeddings + pgvector
- Real-time performance metrics (latency, uptime)
- Side-by-side server comparison
- 24 servers indexed, 91 capabilities tagged
Tech: TypeScript, Supabase, Railway, MCP SDK
Live: https://mcp-discovery-production.up.railway.app
GitHub: https://github.com/yksanjo/mcp-discovery
Feedback welcome!
1 comments