Crow-CLI/crow-CLI: Minimal MCP based ACP agent (github.com)

The Crow-CLI project has introduced a novel two-layer system designed to enhance the interaction between AI models and execution environments. It consists of an Agent Communication Protocol (ACP) agent, known as crow-cli, which handles decision-making, and a Minimal Command Processing (MCP) server, called crow-mcp, that performs tasks based on those decisions. This architecture leverages a React Loop with OpenAI-compatible language models, facilitating seamless tool dispatch and execution across various environments, including local terminals and web-based tools. The significance of this development lies in its innovative agent architecture, which streamlines the integration of AI models with real-world application scenarios while effectively managing resources like session data and tool execution. Key technical features include a PTY-backed terminal for real-time command execution, file handling capabilities, and a sophisticated editing tool utilizing multiple fuzzy matching techniques. The system also supports runtime model switching, cancellation at multiple levels, and employs a SQLAlchemy-backed approach for session management, enhancing user experience and flexibility in automating complex tasks and workflows. This setup paves the way for more robust AI applications that can engage in extended dialogue while managing resource constraints efficiently.