mcp-agent

Source: https://github.com/lastmile-ai/mcp-agent

Category: mcp-middleware-orchestration

Tags: ai-agent, middleware, mcp, open-source

mcp-agent is an open-source, Python-based framework for building effective AI agents using the Model Context Protocol (MCP). It provides simple, composable patterns for constructing robust, production-ready agent applications that can interact with various MCP servers and expose their tools to large language models (LLMs).

Features

• Composable Agent Framework: Build AI agents by chaining together simple workflow patterns. Agents can access a set of MCP servers and expose their tools to LLMs.
• MCP Server Management: Handles the lifecycle of MCP server connections, simplifying integration and orchestration.
• Workflow Patterns: Implements all workflow patterns from Anthropic's "Building Effective Agents" (e.g., Augmented LLM, Parallel, Router, Intent-Classifier, Orchestrator-Workers, Evaluator-Optimizer) and OpenAI's Swarm pattern in a model-agnostic way.
• Multi-Agent Orchestration: Supports collaborative multi-agent workflows and human-in-the-loop operations.
• Augmented LLMs: Enhance LLMs with tools and memory from connected MCP servers. Supports both OpenAI and Anthropic models.
• Router and Intent Classification: Route tasks to the most relevant agents or servers, using embedding or LLM-based classification.
• Evaluator-Optimizer Loops: Implement iterative refinement and quality control using separate evaluator and optimizer agents.
• Orchestrator-Workers Pattern: Automatically parallelizes steps and manages dependencies between agent tasks.
• Swarm Pattern: Reference implementation for multi-agent orchestration, model-agnostic and composable with other workflows.
• Human Input and Signaling: Pause/resume workflows for human input or external signals, supporting human-in-the-loop tasks.
• App Configuration: Flexible YAML-based configuration for servers, logging, execution, and LLM providers.
• Persistent and Managed Server Connections: Fine-grained control over server connection lifecycles, supporting both ephemeral and persistent connections.
• MCP Aggregator: Aggregate multiple MCP servers into a single interface, exposing all tools to LLMs with optional namespacing.
• Example Applications: Includes ready-to-use examples such as Gmail agent, RAG chatbot, Swarm-based customer support, Streamlit and Marimo integrations.
• Open Source and Extensible: Contributions are welcomed. Licensed under Apache-2.0.

Example Use Cases

• Multi-agent collaborative workflows
• Human-in-the-loop processes
• Retrieval-Augmented Generation (RAG) pipelines
• Automated evaluation and optimization loops
• Custom AI agent applications with file, web, or email access via MCP servers

Pricing

mcp-agent is open source and free to use under the Apache-2.0 license.

License

Apache-2.0

Resources

• GitHub Repository
• Documentation & Examples

Note: mcp-agent is under active development and welcomes contributions from the community.