Claude MCP Agent

What is Claude MCP Agent?

Claude MCP Agent is a Python client library that enables AI agents to interact with MCP (Model Context Protocol) servers using structured state management. It accepts any Pydantic model as its state, translating it into context that Claude can understand. The library is designed for developers building agent workflows with LangGraph or similar frameworks.

How to use Claude MCP Agent?

Install the library via pip (not yet published). Configure MCP servers either through a mcp_config.json file or programmatically using ServerConfig objects. Create an MCPAgent instance, optionally selecting a subset of servers, then call process_query() with a query string and state to interact with the servers.

Key features of Claude MCP Agent

• Pure client-side MCP implementation for clean separation of concerns
• Global server configuration and flexible server selection
• Stateful operation with Pydantic models for type‑safe context
• Seamless integration with LangGraph and similar frameworks
• Mixed configuration (JSON file + programmatic server addition)

Use cases of Claude MCP Agent

• Weather analysis and trip planning with structured conversation state
• Building conversational agents that maintain interaction history
• Integrating multiple MCP servers (weather, database, etc.) in one agent
• Creating type‑safe, stateful workflows with LangGraph

FAQ from Claude MCP Agent

What is the difference between Claude MCP Agent and a basic MCP client?

Claude MCP Agent adds structured state management using Pydantic models, translating them into Claude‑compatible prompts. It also supports global server configuration and the ability to use all servers or a specific subset.

How does state management work?

The agent accepts any Pydantic model as its state. This state is updated after each process_query() call, enabling conversation history and context to persist across interactions.

Can I use multiple MCP servers at once?

Yes. You can configure multiple servers in a mcp_config.json file or programmatically, and choose to use all of them or only a specific subset when creating an MCPAgent instance.

Which frameworks and models does it support?

Claude MCP Agent is designed to work with LangGraph and similar agent frameworks. It builds on the Model Context Protocol, which is model‑agnostic, but its state translation is tailored for Claude.

How do I configure servers?

Servers can be configured via a mcp_config.json file (with optional environment variable references) or programmatically using ServerConfig objects. You can also mix both approaches by loading from file and adding additional servers in code.