Repolode
Submit
evalstate

evalstate/fast-agent

Python framework and CLI for building, running and evaluating LLM agents with MCP, Agent Skills, ACP and A2A support.

View on GitHubWebsite
Stars
3,943
Forks
450
Open issues
41
Last commit
6 hours ago

What is it?

fast-agent is a Python framework and command-line tool for working with LLM agents. Agents and workflows are defined in simple prompt and configuration files, and can be chatted with before, during and after a workflow runs to tune and diagnose behavior. It also includes an interactive terminal shell where you can run shell commands alongside the model.

The project connects agents to MCP servers over stdio or Streamable HTTP, including OAuth for remote servers, and supports Agent Skills defined as SKILL.md files. Model support covers Anthropic, OpenAI and Google natively, with Azure, Ollama and many others available through TensorZero. Agents can also be exposed as ACP clients or MCP servers.

For composition, fast-agent provides chain, parallel, evaluator-optimizer, router, orchestrator and agents-as-tools workflows, plus Python function tools and a MAKER pattern for repeated sampling with voting.

Key features

  • 01Interactive CLI shell with shell command support and skill management
  • 02Connects to MCP servers over stdio or Streamable HTTP, with OAuth
  • 03Supports Agent Skills (SKILL.md) and ACP clients
  • 04Declarative chain, parallel, router, orchestrator and evaluator-optimizer workflows
  • 05Native providers for Anthropic, OpenAI and Google, plus Ollama and others via TensorZero

Use cases

  • Running a coding agent with shell access against a local project
  • Prototyping multi-agent workflows in Python with declarative decorators
  • Testing how an LLM interacts with a custom MCP server and diagnosing its transport
  • Exposing a configured agent as an MCP server or ACP client endpoint

Quick start

  1. 1

    Install the uv package manager

  2. 2

    Start an interactive session with shell support

    uvx fast-agent-mcp@latest -x

Things to know

  • Some MCP servers, such as Filesystem and Docker, need configuration changes on Windows.
  • Automation and server runs must set a model via --model, FAST_AGENT_MODEL, default_model, or the agent definition.

Topics

#ai#agent-framework#mcp#llm#cli#python#agent-skills#a2a#acp#agent

FAQ

What is fast-agent?
Python framework and CLI for building, running and evaluating LLM agents with MCP, Agent Skills, ACP and A2A support. fast-agent is a Python framework and command-line tool for working with LLM agents. Agents and workflows are defined in simple prompt and configuration files, and can be chatted with before, during and after a workflow runs to tune and diagnose behavior. It also includes an interactive terminal shell where you can run shell commands alongside the model.
Is fast-agent free to use? What license does it have?
fast-agent is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is fast-agent still maintained?
Actively maintained. The last commit to fast-agent was on Oct 20, 2026.
What can I use fast-agent for?
Running a coding agent with shell access against a local project Prototyping multi-agent workflows in Python with declarative decorators Testing how an LLM interacts with a custom MCP server and diagnosing its transport Exposing a configured agent as an MCP server or ACP client endpoint
How do I install fast-agent?
Install the uv package manager Start an interactive session with shell support: uvx fast-agent-mcp@latest -x
Where can I get fast-agent?
The source code and installation instructions are on GitHub: https://github.com/evalstate/fast-agent

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects