MCP Documentation

GitMem MCP (Manhattan) – Persistent AI memory that reduces token costs by 37%

MCP
GitMem MCP (Manhattan)
Model Context Protocol Server

Overview

AI-powered persistent memory for Claude, Cursor, Copilot, Windsurf, and other AI agents.

Token Savings
37%
Protocol
MCP v1
Transport
SSE / stdio
Tools
17+

What is GitMem MCP?

GitMem MCP is a Model Context Protocol server that gives AI coding agents persistent, semantic memory across conversations. Instead of re-explaining your codebase, preferences, and project context every session, GitMem remembers it — cutting token usage by up to 37%.

It works with any MCP-compatible client: Claude Desktop, Cursor IDE, GitHub Copilot, Windsurf, Cline, and more.

How We Reduce Tokens by 37%

Context Caching

Stores project context once and retrieves only what's relevant, eliminating redundant context in every prompt.

Semantic Dedup

Detects duplicate information across sessions and merges it, preventing the same facts from consuming tokens repeatedly.

Smart Retrieval

Uses vector similarity to inject only the most relevant memories, keeping prompt size minimal and focused.

Key Capabilities

Persistent Memory

Store and retrieve information across conversations and sessions.

Semantic Search

Find relevant memories using natural language queries.

AI-Generated Answers

Get comprehensive answers from memory context directly.

Multi-Agent Support

Separate memory spaces for different projects or use cases.

Analytics & Metrics

Track memory usage, queries, and performance metrics.

Export / Import

Easily export and import your memory data.