MCP Memory Keeper

MCP Memory Keeper is an MCP server that provides persistent context management and memory storage for Claude AI coding assistants. It connects Claude Code or Claude Desktop directly to a local SQLite database stored on the user's machine, eliminating context loss caused by conversation compaction or session resets. Software engineers, DevOps developers, and teams use it to retain important architectural decisions, implementation milestones, code investigation notes, and test results across independent sessions. The server allows AI assistants to organize context using custom categories, priority levels, and topic-based channels that can correlate directly with git branches. It features full-text search, snapshot checkpoints, diff-based change tracking, file content caching, and relationship linking between stored items. By functioning as a persistent shared board, MCP Memory Keeper allows multiple Claude instances to operate concurrently, reading and updating a unified project memory store without overflowing token limits or requiring manual re-prompting.

Category: AI Memory & Context

Tags: context, memory, persistence, sqlite

Visit MCP Memory Keeper

How to install and configure MCP Memory Keeper

1. Open your terminal and add the server to Claude Code using NPX by running: bash claude mcp add memory-keeper npx mcp-memory-keeper 2. Alternatively, to configure it inside Claude Desktop or a project configuration file, add the following entry to your mcpServers block: json { "mcpServers": { "memory-keeper": { "command": "npx", "args": ["mcp-memory-keeper"], "env": { "TOOL_PROFILE": "standard" } } } } 3. Optionally set environment variables such as DATA_DIR to customize the SQLite database path or TOOL_PROFILE (minimal, standard, or full) to limit available tools. 4. Verify your configuration by executing claude mcp list or claude mcp get memory-keeper.

What you can do with MCP Memory Keeper

• Creating snapshot checkpoints during long refactoring sessions to restore complete architectural context after automated compaction resets. - Sharing implementation context and research findings in parallel across multiple concurrent Claude Code terminal sessions. - Tracking architectural decisions and verification results across git branches using topic-based persistent memory channels. - Searching and retrieving historical bug investigations, file cache changes, or previous code designs using full-text SQLite queries. - Exporting and importing persistent assistant project memory across development environments or among team members.

Key facts

• Open Source
• https://github.com/mkreyman/mcp-memory-keeper
• AI Memory & Context, Databases & Data Stores
• context, memory, persistence, sqlite

Part of MCP Servers

How do I install MCP Memory Keeper?

You can install it quickly using the Claude CLI by running claude mcp add memory-keeper npx mcp-memory-keeper. Alternatively, you can install it globally via npm using npm install -g mcp-memory-keeper, or clone the repository and build it locally with npm install and npm run build before adding it.

What does MCP Memory Keeper do?

MCP Memory Keeper provides a persistent local SQLite storage layer for Claude coding assistants. It saves notes, architectural decisions, and progress across context window compactions and terminal restarts, allowing models to organize context using channels, snapshots, categories, and full-text search.

Where is the context data stored?

All context data is stored locally on your machine in a SQLite database located at ~/mcp-data/memory-keeper/ by default. You can change this storage location by setting the DATA_DIR environment variable in your configuration.

Which MCP clients work with MCP Memory Keeper?

MCP Memory Keeper works with Claude Code CLI and Claude Desktop, as well as any other client implementing the standard Model Context Protocol. You can configure it project-wide, globally per user, or directly inside client configuration files.

Is MCP Memory Keeper open source?

Yes, MCP Memory Keeper is open source software distributed under the permissive MIT license. You can inspect its source code, modify it, or build it locally directly from its GitHub repository.