MCP Agent

An open-source Python AI agent framework connects LLMs to MCP tools, combines reusable workflows, and supports pause and recovery through Temporal.

MCP Agent is a Python framework for developers building AI agents that use tools through Model Context Protocol (MCP). It handles connections to MCP servers and provides reusable workflows for coordinating agents across a task.

MCP tool access Connect agents to filesystem, web fetch, Slack, Jira, and other MCP servers, with support for resources, prompts, and OAuth authentication.
Agent workflows Combine routing, map-reduce, orchestration, and evaluation workflows to distribute tasks and review results, using Python code to control branching and repetition.
Pause and recovery Use Temporal-backed execution for workflows that need retries, persistent history, or pauses while waiting for human input or external signals.
Agents as servers Expose agent tools and workflows through an MCP server that Claude Desktop, Cursor, or custom clients can call, or bring several MCP servers behind one interface.
Outputs and monitoring Generate structured responses, retain conversational context, and track execution through structured logs, OpenTelemetry tracing, and token accounting.

The framework is open source under Apache 2.0 and runs as an installable Python application on your own machine or server. It supports OpenAI, Anthropic, Google, Azure, and Amazon Bedrock providers. Using these cloud providers requires credentials and an internet connection, and sends model requests to the chosen service. Temporal is the backend required for durable execution.