MCP Agent is an open-source Python framework for building AI agents entirely on the Model Context Protocol (MCP). Its premise is that MCP is all you need: any tool or data source exposed as an MCP server becomes something your agent can use, and agent behavior is composed from the simple workflow patterns described in Anthropic’s “Building Effective Agents” research. The project has about 8,600 stars on GitHub and was actively developed into 2026.
The framework fully implements MCP — tools, resources, prompts, notifications, OAuth, sampling and more — and manages the lifecycle of server connections for you. On top of that runtime it implements the canonical agent patterns (prompt chaining, routing, parallelization, orchestrator-workers and evaluator-optimizer) as composable building blocks, plus an “augmented LLM” abstraction that works across providers. For production use, workflows can run on Temporal for durable execution that survives crashes and restarts, agents can themselves be exposed as MCP servers, and a managed cloud runtime is available for deployment. The framework also takes care of the operational details that agent demos usually skip: structured logging and OpenTelemetry tracing, human-in-the-loop signals that let a workflow pause for approval, token accounting, and secrets management. A large gallery of examples — from simple finder agents to research and customer-support workflows — shows how the patterns compose, and the project maintains its own documentation site with guides for taking agents from a laptop to production.
MCP Agent is free and open source under the Apache-2.0 license for self-hosting. The team also offers a managed cloud runtime (mcp-c) for hosted deployments, which is the paid side of the project.
MCP Agent is for Python developers who want to build agents the MCP-native way — composing proven patterns instead of wrestling with framework sprawl — and scale them to durable production workflows.