lastmile-ai/mcp-agent

uv add mcp-agent
pip install mcp-agent
uv add "mcp-agent[openai, anthropic, google, azure, bedrock]" (optional LLM providers)

Usage

• •Scaffold: uvx mcp-agent init; configure servers in mcp_agent.config.yaml and secrets in mcp_agent.secrets.yaml
• •Define an MCPApp + Agent(server_names=[...]), attach an AugmentedLLM, call generate_str / generate_structured
• •Deploy: uvx mcp-agent login && uvx mcp-agent deploy my-agent
• •Expose as MCP server: server = create_mcp_server_for_app(app); server.run_stdio()

Key features

• ✓Full MCP support: tools, resources, prompts, notifications, OAuth, sampling, elicitation, roots
• ✓Composable workflow patterns: parallel (map-reduce), router, intent classifier, orchestrator-workers, deep research, evaluator-optimizer, swarm (OpenAI Swarm compatible)
• ✓Durable execution via Temporal (pause/resume/retry/history) without API changes; run a worker alongside the app
• ✓Agent-as-MCP-server (server-of-servers) so MCP clients can call your workflows; FastMCP-compatible server creation
• ✓Observability: structured logging, OpenTelemetry, token counter; signals/human-input; icons; OAuth client

Best for

Python developers building production agents that orchestrate multiple MCP servers with Anthropic's effective-agent patterns and optional Temporal durability.

Caveats

• ⚠It is a framework/SDK/library, not an out-of-the-box runnable MCP server
• ⚠Requires LLM API keys (e.g. OPENAI_API_KEY) in mcp_agent.secrets.yaml or env
• ⚠Cloud deployment is in Beta