Production-Ready FastMCP Server

MCP Server (FastMCP) and Client

Production-ready server and a production-grade client with structured logging, env config, health checks, metrics, and containerization.

Features

• Server: Stdio and SSE runtimes via FastMCP, CORS & security headers, token auth, basic rate limiting

• Client: SSE and stdio transports, CLI to list tools/call tools/get resources, structured logs

• Structured JSON logging with structlog

• Env-based configuration

• Health endpoints and CLI checks

• Prometheus metrics primitives

• Dockerfile and Makefile

Requirements

• Python 3.9+

Setup

Copy and adjust environment:

Run (stdio)

Run (SSE)

Health

Docker

Client CLI

Environment (SSE example):

List tools:

Call tool:

Get resource:

Health check:

Security

• Set a strong AUTH_TOKEN in production for SSE mode

• Restrict CORS_ORIGINS to trusted origins

• Run the container as non-root (Dockerfile does)

• Prefer TLS for SSE (VERIFY_TLS=1)

• Limit client network egress in production and rotate tokens regularly

RAG (Postgres + pgvector)

• Set DATABASE_URL (or PG* envs) and OPENAI_API_KEY.

• Enable vector extension in Postgres (the app will attempt to create it).

Ingest files via CLI:

Ask a question via CLI:

Query with citations via client:

HTTP endpoints (when server running):

• POST /rag/upload (multipart form with files)

• POST /rag/query JSON { "question": "..." }

• GET /rag/chunk/{chunk_id} (get chunk metadata)

MCP tool:

• rag_ask(question: str) -> str

OpenTelemetry Tracing

Enable distributed tracing with:

Traces include:

• RAG ingestion: file parsing, chunking, embedding, DB operations

• RAG retrieval: vector search, BM25 reranking, context assembly

• LLM calls: token usage, model info, latency

• Database operations: SQL queries, connection pooling