概览
What is AI Agent Starter with PydanticAI and MCP?
A base project using PydanticAI’s Agent API that connects to any LLM provider (e.g., OpenAI, Groq, Azure OpenAI) and supports multiple MCP servers via a simple JSON configuration. It includes a minimal demo client to show tool calls and results. Aimed at developers building AI agents with flexible backends.
How to use AI Agent Starter with PydanticAI and MCP?
Clone the repository, copy .env.example to .env, edit with your provider and API key, optionally configure MCP servers in mcp_config.json, then start the API server with Docker (docker-compose up --build) or with UV (uv sync then uv run -- uvicorn src.server.server:app). Access the running server at http://localhost:8000.
Key features of AI Agent Starter with PydanticAI and MCP
- Multi-provider LLM support via
provider:modelsyntax - Connects to multiple MCP servers configured in
mcp_config.json - Demo client included for testing tool calls
- Deployable with Docker or UV
- Azure OpenAI configuration supported
Use cases of AI Agent Starter with PydanticAI and MCP
- Build AI agents that use different LLM providers for the same tool chain
- Integrate several MCP servers (weather, filesystem, etc.) into one agent
- Prototype agent workflows locally with a built-in demo client
FAQ from AI Agent Starter with PydanticAI and MCP
What prerequisites are needed to run the project?
Docker or UV (Python package manager) must be installed. The project also requires API keys for the chosen LLM provider.
How do I configure multiple MCP servers?
Edit the mcp_config.json file at the project root, following the structure from the MCP protocol quickstart. Each entry specifies a server command and its arguments.
Can I use Azure OpenAI instead of standard OpenAI?
Yes. Fill out the additional Azure fields in your .env file as described in the example and the PydanticAI docs.
How can I test custom or third-party MCP servers?
Use the MCP Inspector tool (npx @modelcontextprotocol/inspector) and pass the server command (e.g., uvx mcp-server-fetch) after the -- separator. Ensure the required runtime and dependencies are available in your environment.
Does the server require a constant internet connection?
The server communicates with external LLM providers via API calls, so internet access is needed for those requests. MCP servers may run locally or remotely depending on their configuration.
