Task Memory for AI Agents That Keep Starting Over

What it does Shrimp Task Manager is an MCP server that plugs into AI coding agents like Claude Code or Cline. It turns natural-language requests into structured development tasks with dependency tracking, then stores that state locally so your agent remembers what it was doing even after the context window rolls over.

The interesting bit Rather than treating the LLM as a stateless oracle, Shrimp treats it like a forgetful collaborator who needs a project manager. It enforces planning before execution, supports a reflect command for iterative review, and lets you assign specialized AI agents to specific tasks.

Key highlights

• Persists task history and progress across sessions via local file storage, bypassing token limits
• Decomposes requests into atomic subtasks with automatic dependency tracking
• Includes a research mode for systematic technology evaluation before implementation
• Maintains project-wide coding standards through configurable rule sets
• Optional React-based Task Viewer provides drag-and-drop visual management of the task graph

Verdict Worth a look if you already use MCP-compatible AI clients and are tired of re-explaining project context every session. Skip it if you want a standalone task manager for human teams—this is strictly agent infrastructure.