evalstate/fast-agent

fast-agent is a Python framework for building, running, and evaluating applications that use large language models. It is designed around a command-line interface with an optional interactive terminal, and it positions itself as a coding assistant, workflow platform, and evaluation tool all in one package. The simplest way to try it is a single terminal command using uv, a Python package runner, which downloads and starts fast-agent without a permanent install. Once running, you can chat with a model, run shell commands by prefixing them with an exclamation mark, attach remote tool servers using the slash-connect command, and install preconfigured capability packs designed for tasks like data analysis or coding. The tool supports local models through Ollama and llama.cpp in addition to hosted providers. On the developer side, defining an agent in Python takes a few lines. You attach a decorator to a function with an instruction string, then call the agent with text input or launch an interactive session. Multiple agents can be composed into workflows using patterns described in the README, such as chains where one agent passes output to the next, or networks where a coordinator agent delegates to specialist agents. Configuration files keep prompts and settings out of the main application code, which makes it easier to version-control and adjust behavior without editing Python. The project claims to be the first framework with complete end-to-end tested support for MCP, the Model Context Protocol that standardizes how AI models connect to external tools and data sources. It also supports MCP Elicitations, which allow a model to ask a human for additional input mid-task. Transport diagnostics are built in, including inspection of Streamable HTTP transport traffic, and OAuth is supported for connecting to secured MCP servers. Model support covers Anthropic, OpenAI, Google, Azure, Deepseek, Ollama, and others. The package is distributed on PyPI and documented on the fast-agent.ai website.