Handoff-MCP

LLM context windows reset every session, causing agents to lose track of progress, pending tasks, and architectural decisions. This skill provides a durable memory layer for your agent using a project-local MCP server to ensure continuity between sessions.

What it does

• Persistent TODO tracking manages concrete tasks with priorities, tags, and status updates directly in the project scope.
• Session handoffs record the current state, key decisions, and specific file paths at the end of a work interval.
• Context management generates reports on active context and compacts bloated sessions into digestible summaries.
• Durable storage keeps all breadcrumbs and tasks in a local store independent of the conversation history.

How it works

1. Use handoff_list and todo_list at the start of a new session to restore project state.
2. Add new tasks via todo_add as requirements emerge during development.
3. Call context_report when the prompt window feels cluttered to identify what to prune.
4. Execute handoff_add before closing the session to leave instructions for the next agent instance.

Frameworks & tools

This skill requires the Handoff MCP server. It is designed for use with Model Context Protocol compatible environments like Claude Desktop, Cursor, or specialized IDE agents.

Why this beats prompting it yourself

Relying on chat history for project state leads to hallucinations as the context window fills up. This skill moves project tracking into a structured local database that persists even if you clear your chat or switch models.

Use cases

• Resuming a complex refactoring job after a break without re-explaining the plan.
• Coordinating between multiple specialized agents by leaving breadcrumbs in a shared store.
• Managing long-running project roadmaps directly within the developer environment.

Known limitations

Requires a running Handoff MCP server. Storage is repository-scoped and does not synchronize across different machines unless the local store is shared.