[Kaap10]

Summary

Closes #65

Adds a FastMCP server implementation for dynavec, enabling AI assistants (Claude Desktop, Cursor, Antigravity, and custom agent runtimes) to perform semantic search and Knowledge Graph RAG over the Model Context Protocol.

Key Features

• FastMCP Server (dynavec.mcp): Exposes dynavec_search (with filtering, rescoring, reranking) and dynavec_graph_search (with seed entities, hops, and vector ranking) tools.
• Environment-Based Configuration: client_from_env() automatically parses DYNAVEC_* and AWS_* variables and configures embedders (OpenAI, Gemini, Voyage, Mistral, Bedrock, Sentence Transformers) and credentials.
• CLI Entry Point: Added dynavec mcp [--transport stdio|sse] [--port PORT].
• Lightweight & Isolated: Lazy imports ensure the core package remains lean (boto3 + numpy only) unless mcp extra is used.
• Unit Tests & Docs: Added unit tests in tests/test_mcp_server.py, example configuration in examples/mcp_server_config.json, and updated documentation.

Please let me know if any changes are needed!

[codeforstartups]

Verified locally (153 passed, ruff green). FastMCP server exposing search + graph_search as MCP tools, env-driven config with embedder auto-detection, CLI + docs + example config, and a 231-line test. Implements #65 (dynavec MCP server) — a roadmap headliner. Thanks @Kaap10!