This MCP server provides a Python FastAPI backend that runs a LangGraph Deep Agents workflow integrated with CopilotKit for a real-time generative UI experience. The backend exposes a LangGraph agent at the root endpoint and a health check at /healthz, while the frontend (Next.js with CopilotKit) communicates with the backend to render tool results as dynamic UI components (e.g., weather cards) as the agent calls tools like get_weather. To use it, start the backend service, then run the frontend, and interact with the UI to trigger tool calls and streaming agent responses. The Deep Agents framework powers multi-step task planning, tool usage, memory checkpoints, and streaming responses, enabling an interactive weather assistant demonstration with real-time UI updates.

Prerequisites:

• Python 3.13 or compatible environment
• uv (uvicorn-based tooling) for running the backend in this project’s setup
• Node.js and npm/yarn for the Next.js frontend

1. Install system dependencies (example for macOS/Linux):

   • sudo apt-get update && sudo apt-get install -y python3.13 python3-venv npm

   • Or use your preferred Python/Node version manager

2. Set up the backend (FastAPI + Deep Agents):

   • cd backend
   • python -m venv venv
   • source venv/bin/activate # on Windows use venv\Scripts\activate
   • pip install -U pip

   • Install project dependencies from pyproject.toml / requirements.txt as appropriate

   • uv sync

3. Run the backend locally:

   • uv run python src/agent/main.py
   • The server will listen on http://0.0.0.0:8123

4. Set up and run the frontend (Next.js with CopilotKit):

   • cd frontend
   • npm install
   • npm run dev
   • The frontend will typically run on http://localhost:3000

5. Optional: Build a Docker image (if a Dockerfile is provided):

   • make build # if a Makefile is configured for building the backend image

6. Run tests (if configured):

   • uv run pytest

Note: The project structure indicates a backend under backend/ and a frontend under frontend/; ensure you install dependencies for both and configure any environment variables required by CopilotKit and LangGraph integrations.

Tips and caveats:

• The backendroot endpoint (/) serves the LangGraph AGUI agent, while /healthz is a dedicated health check.
• If you modify tools (e.g., add a get_weather tool), ensure the Deep Agents graph is rebuilt and the FastAPI endpoint is restarted to reflect changes.
• Environment variables like PORT can be adjusted to fit your deployment; the MCP config above exposes PORT=8123 for uvx usage.
• When developing locally, enable hot-reloading in the frontend (CopilotKit) and ensure the frontend can reach http://127.0.0.1:8123 for the backend API.
• If you encounter memory or state issues, leverage the Memory Checkpointer in the Deep Agents setup to persist conversation state between sessions.
• For production runs, consider containerizing both backend and frontend with proper ingress/egress rules and using a Kubernetes manifest similar to the project’s Skaffold setup.