mcp-agent

Stale
GitHub Python Apache-2.0

Description

mcp-agent is LastMile AI's toolkit for building agents around Model Context Protocol integrations, making it easier to connect MCP tools into multi-step workflows.

Key Features

  • Full MCP implementation with automatic lifecycle management of MCP server connections
  • Composable agent patterns from Anthropic's Building Effective Agents guide (map-reduce, orchestrator, evaluator-optimizer, router, etc.)
  • Durable execution via Temporal runtime — pause, resume, and recover workflows without API changes
  • Create MCP servers with FastMCP-compatible API, expose agents as MCP services
  • CLI scaffolding with `uvx mcp-agent init` and one-command cloud deployment
  • Complete support for Tools, Resources, Prompts, OAuth, Sampling, Elicitation, and Roots

Use Cases

💡 Building multi-step agent workflows that connect multiple MCP tools
💡 Integrating MCP tools into production-grade agent applications
💡 Creating durable agent tasks that can be paused, resumed, and recovered
💡 Building MCP servers usable from ChatGPT and other clients
💡 Rapidly prototyping Anthropic's effective agent patterns

Tags

mcp agent tools python

Categories

🔌 MCP Tools

Quick Start

Install: pip install mcp-agent or scaffold with `uvx mcp-agent init`. Configure your LLM API key in mcp_agent.secrets.yaml. Define an Agent with server_names and instruction, attach an AugmentedLLM to interact. Deploy with `uvx mcp-agent deploy` to cloud.
Visit GitHub Visit Website View Docs

Related Projects