Chat Agent

Central chat API powered by the OpenAI Agents SDK with MCP (Model Context Protocol) support. Accepts authenticated requests, keeps per-session conversation history in Redis, and can connect to one or more MCP servers (e.g. finance-app) so the agent can use remote tools in conversation.

Features

• Chat API: POST with session_id and message; returns agent reply
• API key authentication via x-api-key header
• Rate limiting: 20 requests per minute per client (SlowAPI)
• Session memory: Redis-backed conversation history (24h TTL) for multi-turn context
• MCP integration: Optional comma-separated list of MCP server SSE URLs; agent connects at first use and can call tools from those services
• Docker Compose with Redis; runs on a shared web network for Caddy and upstream MCP services

Prerequisites

• Docker and Docker Compose
• An existing Docker network named web (e.g. docker network create web)
• OpenAI API key (for the agent)

Quick Start

1. Clone and enter the repo

   git clone https://github.com/youssefaltai/chat-agent.git
   cd chat-agent

2. Configure environment

   cp .env.example .env

   Edit .env and set:

   • OPENAI_API_KEY — your OpenAI API key (required for the agent)
   • API_KEY — secret for the chat API (clients send this in x-api-key header; change from default)
   • REDIS_URL — Redis connection URL (default: redis://redis:6379/0 for Docker)
   • MCP_SERVER_URLS — optional; comma-separated MCP SSE URLs (e.g. http://finance_app:3000/mcp/sse)

3. Run with Docker

   docker compose up -d --build

   The service runs as the siri_agent container and depends on redis (container name chat_redis).

4. Verify

   • GET http://localhost:3000/health → {"status": "ok"}
   • Chat: POST http://localhost:3000/ with header x-api-key: <API_KEY> and body {"session_id": "s1", "message": "Hello"}

Configuration

Variable	Description	Default
OPENAI_API_KEY	OpenAI API key for the agent	— (required)
API_KEY	Secret for x-api-key (chat API auth)	change-me
REDIS_URL	Redis URL for session history	redis://redis:6379/0
MCP_SERVER_URLS	Comma-separated MCP SSE URLs	— (optional)

API

• GET /health — Health check; returns {"status": "ok"}.
• POST / — Chat. Requires header x-api-key and JSON body:

  { "session_id": "string", "message": "string" }

  Returns:

  { "reply": "string" }

  Rate limit: 20 requests per minute per IP.

Project Structure

app/
  api.py      # FastAPI app: health, chat endpoint, API key auth, rate limiting
  agent.py    # OpenAI Agents SDK agent; lazy init, MCP server connections
  chat.py     # Chat handler: load history, run agent, persist turn
  memory.py   # Redis-backed session history (get_history, append_message)
  config.py   # Pydantic settings from .env

Integration

• Caddy: Add a route, e.g. handle_path /chat/* { reverse_proxy siri_agent:3000 } (adjust port if you change the app port).
• MCP services: Run finance-app (or other MCP services) on the same web network and set MCP_SERVER_URLS so the agent can use their tools.

License

MIT