🚀 FastAPI AI-Augmented Architecture Boilerplate

A production-ready, out-of-the-box template for building asynchronous AI microservices using Python 3.12, FastAPI, and the Model Context Protocol (MCP).

⚡ Core Features

• Asynchronous API: Built on FastAPI and Uvicorn for maximum throughput.

• Conversational AI Agent: Pre-configured with OpenAI and SQLite for built-in memory and context persistence.

• Model Context Protocol (MCP): Exposes a native SSE server for external AI clients (Cursor, Claude Desktop) to autonomously execute backend tools.

• Containerized Infrastructure: Fully Dockerized for zero-friction local development and deployment.

Related MCP server: Streamable HTTP MCP Server

🚀 Quick Start

1. Configure the Environment Create a .env file in the root directory and add your AI provider API key:

echo "OPENAI_API_KEY=sk-your-key-here" > .env

2. Spin up the Infrastructure

docker-compose up -d --build

The FastAPI server will be available at http://localhost:8000.

🧪 Testing the Architecture

Once your Docker containers are running, you can immediately verify both the AI memory persistence and the MCP tool execution.

1. Testing the Conversational API

We will test if the AI can remember context using the SQLite database.

Step A: Initiate the Conversation

curl -X POST http://localhost:8000/api/chat \
     -H "Content-Type: application/json" \
     -d '{"message": "My favorite framework is FastAPI."}'

Note the conversation_id returned in the JSON response.

Step B: Test Context Memory Pass the ID back to the server to verify database persistence:

curl -X POST http://localhost:8000/api/chat \
     -H "Content-Type: application/json" \
     -d '{
           "message": "What did I say my favorite framework was?",
           "conversation_id": "PASTE_YOUR_ID_HERE"
         }'

2. Testing the MCP Server (System Health Tool)

The most professional way to test an MCP Context Provider without an IDE is using the official open-source MCP Inspector.

Step A: Run the Inspector Run this command in your local terminal (requires Node.js):

npx @modelcontextprotocol/inspector http://localhost:8000/mcp/sse

Step B: Execute the Tool

1. The inspector will open a debugging dashboard in your browser.

2. Navigate to the Tools tab.

3. You will see the system_health tool automatically listed via the server's handshake.

4. Click Execute to watch the MCP server query the Docker environment and return the Python version and SQLite health status in real-time.

🔌 Connecting to Cursor / Claude Desktop

To utilize this backend as a context provider in your daily development:

1. Open your AI client's MCP configuration settings.

2. Add a new MCP server connection.

3. Select the SSE (Server-Sent Events) transport method.

4. Set the URL to: http://localhost:8000/mcp/sse