// the find
NicholasGoh/fastapi-mcp-langgraph-template
A modern template for agentic orchestration — built for rapid iteration and scalable deployment using highly customizable, community-supported tools like MCP, LangGraph, and more.
A FastAPI + LangGraph starter aimed at teams building agentic backends that talk to MCP servers, with SQLModel/Postgres, Supabase for RBAC, and Langfuse for LLM tracing wired in. It's for someone who wants a batteries-included skeleton for an agent API rather than building the FastAPI/LangGraph/MCP plumbing from scratch.
The MCP architecture is concrete, not hand-wavy — there's an actual multi-server setup (DBHub, a custom Youtube MCP server, an Inspector) with a documented SSE topology, and a build script that compiles the Youtube MCP image inside a throwaway Docker-in-Docker container so it doesn't leave build artifacts on your machine. Dev/prod are properly split via separate compose files, and there's a working VSCode devcontainer plus an entrypoint override for attaching the FastAPI debugger, which is more debugging setup than most templates bother with. LangGraph gives you streaming and checkpointed persistence out of the box (there's a dedicated `checkpoints.py` router), which is the annoying part of agent orchestration to get right yourself.
A large chunk of what's advertised in the topics and stack list — Prometheus, Grafana, Auth0/OIDC, CI/CD, Terraform, ECR/Fargate deploy — is explicitly marked 'Planned Features' in the README and doesn't exist in the code yet, so don't adopt this expecting turnkey observability or auth. There's no test directory anywhere in the tree, which is a real gap for something positioned as a production template. The README leans heavily on shields.io badges and star-history charts rather than usage detail, and the actual application logic is thin (one orchestration.py, one persistence.py, three routers) — you're getting scaffolding and Docker/nginx wiring more than a fleshed-out agent implementation.