mcp-agent

✓ Text source-checked Oct 23, 2026

Build effective agents using Model Context Protocol and simple workflow patterns

Source-checked against the vendor's site Oct 23, 2026 · rebuilt from the vendor's own pagesProfile updated Oct 23, 2026Link checked Oct 25, 2026How we rate

FreeMCP ServersProvider-agnostic — works with OpenAI (GPT-4o family), Anthropic (Claude 3.5/3.7), Google (Gemini), Azure OpenAI, and AWS Bedrock (not source-checked)

Visit website →

In short

mcp-agent builds effective AI agents using Model Context Protocol and simple workflow patterns. It supports multiple LLM providers and scales to complex workflows via Temporal.

mcp-agent fully implements MCP and manages the lifecycle of MCP server connections. mcp-agent implements the patterns described in Anthropic's Building Effective Agents in a composable way. mcp-agent supports simple agents and scales to workflows built on Temporal for pausing, resuming, and recovering without API changes.

mcp-agent supports LLM providers including OpenAI, Anthropic, Google, Azure, and Bedrock via optional packages. mcp-agent allows exposing an application as a standard MCP server for clients like Claude Desktop and Cursor.

mcp-agent provides a CLI available via uvx. mcp-agent offers a cloud deployment option for managed Temporal execution, secrets, and HTTPS MCP endpoints.

From the vendor's pages

According to mcp-agent: Mcp-agent fully implements MCP and manages the lifecycle of MCP server connections. Its pages list this strength: Any MCP server connects without custom adapters.

— The AI Tool Bible editorial team

Pros

✅ Any MCP server connects without custom adapters.
✅ The same code annotations work for both asyncio and Temporal execution.
✅ mcp-agent enables structured logging, OpenTelemetry, and programmatic token usage tracking.
✅ mcp-agent creates MCP servers with a FastMCP-compatible API.

Facts

Platforms: cli (not source-checked)
API available: Yes (not source-checked)
Open source: Yes (license: Apache-2.0)
GitHub stars: 8,566 (checked Oct 29, 2026) (not source-checked)
Last GitHub push: Jan 25, 2026 (not source-checked)
First commit: Dec 2024 (not source-checked)
Model used: Provider-agnostic — works with OpenAI (GPT-4o family), Anthropic (Claude 3.5/3.7), Google (Gemini), Azure OpenAI, and AWS Bedrock (not source-checked)
Package downloads (30 d): 25,742 · pypi mcp-agent · checked Oct 15, 2026 (not source-checked)

Where this tool sits

MCP Servers Agent Frameworks MCP frameworks AI research assistant For Developers Has Sdk Open Source Open Source Free Powered By Claude Self Hostable

Use cases

Parallel map-reduce workflowsMulti-agent swarm handoffsDeep research workflows

Frequently asked

How much does mcp-agent cost?: mcp-agent is free. The provided documentation does not list specific pricing tiers or additional costs for cloud deployment features.
Which AI models does mcp-agent support?: It is provider-agnostic, supporting OpenAI (GPT-4o), Anthropic (Claude 3.5/3.7), Google (Gemini), Azure OpenAI, and AWS Bedrock via optional packages.
Can I use mcp-agent with existing tools like Cursor?: Yes, mcp-agent allows exposing an application as a standard MCP server, compatible with clients such as Claude Desktop and Cursor.
Does mcp-agent handle complex workflow management?: Yes, it scales to workflows built on Temporal, enabling pausing, resuming, and recovering processes without requiring API changes.
How do I install or run mcp-agent?: mcp-agent provides a CLI available via uvx. It also offers a cloud deployment option for managed Temporal execution and HTTPS MCP endpoints.