Limps

About

Limps addresses the challenge of context drift among different large language model (LLM) providers by establishing a single source of truth for AI agent planning documents, task status, and decisions. It functions as a Local Intelligent MCP Planning Server, accessible via a standardized Model Context Protocol (MCP) interface. This enables seamless integration with tools like Claude Desktop, Cursor, and GitHub Copilot, allowing agents to search, create, update documents, and claim tasks from a centralized repository. It supports local deployment for privacy or server deployment for global access, facilitating robust AI agent coordination and project management.

Key Features

• Git-based Document Versioning: Supports pointing the server at a Git repository to enable version history, branching, collaboration via pull requests, and backups for all planning documents.
• Standardized MCP Interface: Connects with various LLM tools like Claude Desktop and Cursor for unified agent planning and task management.
• Comprehensive MCP Tools: Offers 16 dedicated tools for document operations (create, update, search), plan management (create, list, status), and task coordination (get next, claim, release, update status).
• CLI for Project & Plan Management: Provides a full command-line interface for initializing projects, starting the MCP server, listing plans and agents, and managing project configurations.
• 0 GitHub stars
• RLM (Recursive Language Model) Support: Implements the RLM pattern from MIT CSAIL for programmatic document examination, recursive processing, and document extraction via a secure QuickJS sandbox.

Use Cases

• Managing AI agent planning documents and tasks across different LLM assistants (e.g., Claude, Cursor, GitHub Copilot) to maintain a shared context.
• Implementing robust development workflows for AI projects, such as managing feature plans, agent coordination, and task tracking across the development lifecycle.
• Version controlling planning documents and decision logs using Git, enabling collaboration, historical tracking, and experimental planning through branching.