Agentpack is an MCP server tool developed by Liqiongyu, built primarily with Rust and distributed via cargo-dist binary releases. This free tool functions as an AI-first local asset control plane, designed to manage and deploy coding-agent assets across various AI coding agent systems and development tools. It offers a declarative and safe approach for integrating assets with platforms like Codex, Claude Code, Cursor, and VSCode. Agentpack helps developers orchestrate and maintain their AI agent configurations effectively, ensuring consistent deployment practices.

What does Agentpack connect to?

Agentpack serves as an AI-first local asset control plane, connecting to various AI coding agent systems and development tools. It integrates with platforms such as Codex, Claude Code, Cursor, and VSCode through explicit target adapters and mappings. This allows it to manage and deploy coding-agent assets consistently across different development environments. Compatible AI clients include Claude Code, OpenAI Codex, and Cursor Agent, making it a central hub for agent deployment.

Core features and tools

The agentpack mcp serve command exposes several callable tools. These include plan, diff, status, and doctor for read-only operations. For state-changing actions, it provides deploy_apply and rollback, which are approval-gated mutations. Agentpack orchestrates these operations using stable automation contracts. While exact transport protocols are not detailed, mentions of MCP server integration suggest HTTP/S as a likely underlying transport for some interactions.

Who is it for?

Agentpack is intended for developers who work with AI coding agents and require a consistent method for managing and deploying agent assets. It supports workflows for deploying instructions, skills, slash commands, and custom prompts across multiple development tools. Developers needing to maintain version control and safe deployment practices for their AI agent configurations will find it useful. Skip this if you do not actively use AI coding agents or require asset deployment across multiple development tools.

How to get started

No credentials are required for agentpack mcp serve commands, as mutations are approval-gated.

Use cases

• Deploying the same agent assets, such as instructions, skills, slash commands, and custom prompts, to various development tools including Codex, Claude Code, Cursor, and VSCode.
• Implementing a preview-first deployment model that includes snapshots and rollback capabilities for safe deletions and changes.
• Executing full end-to-end deployment workflows, encompassing updates, preview diffs, applying deployments, checking status, and rolling back to specific snapshot IDs.

Agentpack provides developers with a structured control plane for managing AI coding agent assets across diverse tools. Its declarative approach and approval-gated mutations for deployments ensure controlled and safe operations. The platform supports key workflows, from multi-tool asset deployment to preview-first models with rollback options. This makes it a valuable utility for maintaining consistent agent configurations. Try deploying agent assets to multiple development tools like Codex and VSCode.