mcp-agent review

Q: How do I install mcp-agent?

Run uv add "mcp-agent". The mcp-agent package needs Python 3.10 or newer. The install section above has the other options from the project's README.

mcp-agent is a Python framework for building AI agents with Model Context Protocol servers and composable workflow patterns.

Updated Nov 1, 2026. Review written Oct 22, 2026 from the project's own sources; usage data refreshed every Monday.

TL;DR

mcp-agent is a Python framework for building agents on Model Context Protocol. It is aimed at Python developers who want agents that connect to MCP servers. It implements the patterns from Anthropic's Building Effective Agents, manages MCP server connection lifecycles, and can run workflows on Temporal for durable execution.

8.6kGitHub stars
+4new stars, week of Oct 25
18.5kdownloads a week, +80% vs 4-week avg
#43in MCP and agent tools this week

mcp-agent at a glance

Category: MCP and agent tools
Maintained by: lastmile ai (organization, @lastmile-ai on GitHub)
License: Apache 2.0
Language: Python, TypeScript, CSS
Repository: github.com/lastmile-ai/mcp-agent
PyPI package: mcp-agent 0.2.6
Latest release: mcp-agent 0.2.6 on PyPI, Dec 5, 2025
First published: Dec 18, 2024

What mcp-agent does

mcp-agent implements MCP and manages the lifecycle of MCP server connections, including tools, resources, prompts, notifications, OAuth, sampling, elicitation and roots.
mcp-agent provides composable workflow patterns such as map-reduce, orchestrator, evaluator-optimizer and router, based on Anthropic's Building Effective Agents.
mcp-agent can run workflows on Temporal, so agents can pause, resume and recover without API changes.
mcp-agent creates MCP servers with a FastMCP-compatible API and can expose agents as MCP servers.
mcp-agent wraps LLM provider SDKs as Augmented LLMs with generate, generate_str and generate_structured methods for structured output.
mcp-agent supports LLM providers including OpenAI, Anthropic, Google, Azure, Bedrock and Ollama through optional packages and Augmented LLM classes.
mcp-agent connects to MCP servers over stdio, SSE and WebSockets, with configurable HTTP headers.

How it works

You write Python code around a single MCPApp runtime. MCPApp loads configuration from mcp_agent.config.yaml, secrets from mcp_agent.secrets.yaml and environment variables, and registers MCP servers and agents. Each agent lists the MCP servers it may call, and you attach an LLM to it. The execution engine is asyncio or Temporal. You can deploy apps yourself or to a managed cloud runtime in beta.

How to install and start using mcp-agent

With uv

uv add "mcp-agent"

With pip

pip install mcp-agent

With OpenAI provider extras

uv add "mcp-agent[openai]"

Requirements

Python 3.10 or later.
An API key for an LLM provider such as OpenAI.
Optional: Temporal for durable workflow execution.

First steps

Scaffold a project with uvx mcp-agent init.
Add your LLM API key to mcp_agent.secrets.yaml or set OPENAI_API_KEY in the environment.
List the MCP servers your agent uses, such as fetch and filesystem, in mcp_agent.config.yaml.
Run the app with uv run main.py.

Who mcp-agent is for

Python developers building agents that call existing MCP servers like filesystem, fetch, Slack or Jira.
Teams that want multi-agent patterns like orchestrator or router without writing custom architecture.
Developers who need long-running agent workflows that can pause, resume and recover through Temporal.
Developers who want to expose their agents as MCP servers.

Limitations to know

mcp-agent is Python-only for agent code. The examples use async Python with asyncio.
mcp-agent cloud deployment and the managed runtime are in beta.
mcp-agent is built around MCP, so tools must be exposed as MCP servers to connect without custom adapters.
mcp-agent is still pre-1.0, at version 0.2.6 on PyPI.

Is mcp-agent growing? Usage data

In the week of Oct 25–31, 2026, mcp-agent ranked #275 on the Devtoolwatch Board and #43 in MCP and agent tools, with a momentum score of 35 out of 100. Its main package had 18.5k downloads that week, +80% against its own 4-week average.

New GitHub stars per week for the last 12 weeks, from GitHub's stargazer history (today's stargazers by the week they starred).

Show the weekly numbers

New GitHub stars per week for mcp-agent
Week of	New stars
Aug 9, 2026	14
Aug 16, 2026	8
Aug 23, 2026	12
Aug 30, 2026	7
Sep 6, 2026	13
Sep 13, 2026	14
Sep 20, 2026	13
Sep 27, 2026	13
Oct 4, 2026	5
Oct 11, 2026	8
Oct 18, 2026	6
Oct 25, 2026	4

On Hacker News

Show HN: Mcp-Agent – Build effective agents with Model Context Protocol, 80 points and 28 comments (Jan 29, 2025)
Show HN: Representing Agents as MCP Servers, 58 points and 16 comments (May 21, 2025)
Show HN: I built an AI agent that helps me invest, 32 points and 23 comments (Jul 19, 2025)

The Board ranks tools by four signals each week: new stars, new stars for the repo's size, added downloads and download growth beyond the market. How the ranking works.

mcp-agent alternatives

Other MCP servers and agent tools on the Board in the week of Oct 25–31, 2026, by momentum:

MCP and agent tools ranked by momentum, the week of Oct 25–31, 2026
Rank	Tool	Stars	Downloads/wk
#1	REAREA is an MCP server and CLI that lets coding agents reverse engineer native binaries, JavaScript and Electron apps, .NET assemblies,…	415.5k	760.6k
#2	ha-mcpha-mcp is an unofficial Model Context Protocol server that lets AI assistants control and manage Home Assistant using natural language.	5.4k	118.1k
#3	TregTreg is a registry and proxy that gives AI agents one token for a catalog of paid third-party API endpoints and team-shared tools.	6.9k	6.7k
#4	Chrome DevTools MCPChrome DevTools MCP is an MCP server and CLI that lets coding agents control and inspect a live Chrome browser using Chrome DevTools.	54.6k	2.7M
#5	codebase-memory-mcpHigh-performance code intelligence MCP server. Indexes codebases into a persistent knowledge graph — average repo in milliseconds. 162…	47.9k	–
#6	open-computer-use👾 Open Computer Use – Open-Source Alternative to Codex Computer Use	2.8k	–

All MCP servers and agent tools, ranked

Recent mcp-agent releases

v0.0.21, May 9, 2025
v0.0.18, May 2, 2025
v0.0.16, Apr 17, 2025
v0.0.15, Apr 16, 2025
v0.0.11, Mar 25, 2025

mcp-agent FAQ

Is mcp-agent free and open source?

Yes. mcp-agent's source code is on GitHub (lastmile-ai/mcp-agent) under the Apache 2.0 license, so you can use it for free.

How do I install mcp-agent?

Run uv add "mcp-agent". The mcp-agent package needs Python 3.10 or newer. The install section above has the other options from the project's README.

Is mcp-agent still maintained?

Activity has slowed: the lastmile-ai/mcp-agent repository was last updated on Jan 25, 2026, and the latest release, mcp-agent 0.2.6 on PyPI, came out on Dec 5, 2025.

How popular is mcp-agent?

mcp-agent has 8.6k GitHub stars as of Nov 1, 2026. It gained 4 new stars in the week of Oct 25–31, 2026, against a 12-week average of 10 a week. Its main package had 18.5k downloads that week, +80% against its own 4-week average. That made it #43 among MCP servers and agent tools on the Devtoolwatch Board.

What are the best alternatives to mcp-agent?

Other MCP servers and agent tools on the Devtoolwatch Board in the week of Oct 25–31, 2026, by momentum: REA (#1), ha-mcp (#2), Treg (#3), Chrome DevTools MCP (#4), codebase-memory-mcp (#5). The alternatives table above has their numbers.

Which LLM providers does mcp-agent support?

mcp-agent supports OpenAI, Anthropic, Google, Azure and Bedrock through optional packages. Release notes also add Ollama, Gemini, Azure OpenAI and OpenAI o4 reasoning model support. You attach a provider by passing an Augmented LLM class, such as OpenAIAugmentedLLM, to an agent.

Can mcp-agent run durable, long-running workflows?

Yes. mcp-agent can use Temporal as the agent runtime backend. Workflows can pause, resume and recover. The same decorators work for both asyncio and Temporal execution, so moving to Temporal needs no API changes.

Can I expose an mcp-agent agent as an MCP server?

Yes. mcp-agent creates MCP servers with a FastMCP-compatible API and can expose agents as MCP servers. Apps deployed to its cloud runtime are all deployed as MCP servers.

Where does mcp-agent store API keys and configuration?

mcp-agent reads settings from mcp_agent.config.yaml, mcp_agent.secrets.yaml, environment variables and optional preload strings. Keep the secrets file out of source control, for example by adding it to .gitignore.

Which MCP transports does mcp-agent support?

mcp-agent supports stdio, SSE and WebSocket connections to MCP servers. You can pass HTTP headers to WebSocket and SSE connections.

Sources

mcp-agent README badge

Maintainers can show the tool's weekly Board rank in their README:

[![Devtoolwatch](https://index.devtoolwatch.com/badge/mcp-agent.svg)](https://index.devtoolwatch.com/tools/mcp-agent/)

How this page is made: Devtoolwatch writes each review from the project's README, releases, package metadata and website, with an AI model, then checks every claim against the source it cites and drops what the sources don't support. The numbers come from Devtoolwatch's daily GitHub snapshots, GitHub's stargazer history, and the npm and PyPI download APIs. Spotted an error? Email [email protected].