This repository provides a template for building agentic orchestration systems, targeting developers who need to rapidly iterate and deploy scalable AI applications. It leverages MCP for standardized LLM context, LangGraph for workflow management, and FastAPI for a robust backend, enabling flexible, provider-agnostic agent development with features like native streaming and persisted state.

How It Works

The architecture centers around an API server built with FastAPI, interacting with various microservices (e.g., DBHub, Youtube) via a defined MCP protocol. MCP standardizes how applications provide context to LLMs, allowing for community integrations and avoiding vendor lock-in. LangGraph manages complex agentic workflows, while Supabase serves as the database backend. Observability is integrated via Langfuse for tracing and Grafana for metrics.

Quick Start & Requirements

• Install/Run: Use docker compose -f compose-dev.yaml up -d for development or docker compose up -d for production.
• Prerequisites: Docker, bash, zsh, OpenAI API key, Supabase credentials, YouTube API key.
• Setup: Build community images with ./community/youtube/build.sh, then build other images with docker compose -f compose-dev.yaml build. Copy and populate environment files (.env.sample to .env, ./envs/backend.env, ./envs/youtube.env).
• Docs: Read the docs with demo videos here

Highlighted Details

• MCP protocol for standardized LLM context and community integrations.
• LangGraph for complex agentic workflow orchestration.
• FastAPI backend with native streaming and persisted chat history.
• Supabase for database and authentication (OIDC & JWT via OAuth 2.0).
• Integrated observability with Langfuse and Grafana.

Maintenance & Community

The project is maintained by NicholasGoh. Further community support details are not explicitly provided in the README.

Licensing & Compatibility

The README does not explicitly state a license. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The template is described as a starting point, and some planned features like LLM observability and SaaS authentication are not yet implemented. The current setup does not connect to all MCP servers.