Codewhale MCP Client: Rust Agent Engine for Multi-Provider Control
Codewhale: A Rust MCP Client for Agent-Driven LLM Orchestration
Codewhale is an open-source Rust-based MCP client designed as an agent engine and terminal client for interacting with a wide array of language models. It provides developers with fine-grained control over AI interactions, particularly for complex, multi-step tasks, emphasizing provider flexibility and operational transparency.
Connecting to Any LLM, Locally or Remotely
A core strength of Codewhale is its extensive model provider support, ensuring developers are not locked into a single ecosystem. It ships with over 40 built-in routes for major providers like Anthropic, DeepSeek, Google, Mistral, Moonshot, OpenAI, OpenRouter, and xAI. Beyond these, Codewhale can connect to any OpenAI-compatible endpoint. For local model execution, it integrates directly with Ollama, vLLM, or SGLang, making it versatile for both cloud-based and on-premise AI development.
Maintaining Control Over Agent Operations
Codewhale prioritizes developer control and transparency throughout the agent's lifecycle. It offers distinct operational modes: Plan mode allows exploration without making changes, while Work and Operate modes facilitate actual modifications. Critical for managing agent behavior are the configurable approval postures, which dictate when a tool call requires explicit developer consent. For accountability and recovery, the client includes /undo and /restore commands for workspace changes, and /receipts to list every file, command, and approval within a session.
Architecting Long-Running AI Jobs
For developers tackling larger, more persistent AI tasks, Codewhale provides several mechanisms to structure and manage work. A durable /goal can be set to guide long-term objectives. Work can be delegated to specialized sub-agents, or developers can run supervised agent teams (FLEET) with pre-spend checks. For repeatable processes, these agent interactions can be scripted as checked-in workflows.
Extending Codewhale with MCP and Plugins
Codewhale's extensibility is key to its utility as an MCP client. It allows direct connection to MCP servers, enabling broader integration within an AI ecosystem. Developers can install skills and plugins to expand its capabilities, and hooks can be run on session and tool events for custom automation. Notably, Codewhale also supports loading existing Claude Code plugins, providing a path for leveraging existing toolsets. An included Computer Use plugin adds tools for observing and operating other applications, though its access requires review and explicit enablement.
References