Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Claude Memory MCP Serversearch for my notes about the project requirements from last week"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Memory MCP
Local, persistent memory for Claude Desktop and MCP-compatible AI assistants.
Part of the
A lightweight MCP server that gives your AI durable, searchable memory — entirely on your machine. Built with TypeScript, SQLite + FTS5, and minimal dependencies.
Quick Start
Choose your installation method below based on your platform and preference.
Option 1: Direct from GitHub (Always Latest)
This method fetches directly from GitHub, bypassing npm cache issues.
macOS / Linux:
Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS, ~/.config/Claude/claude_desktop_config.json on Linux):
Windows (Command Prompt wrapper):
Add to your Claude Desktop config (%APPDATA%/Claude/claude_desktop_config.json):
Windows (Full npx.cmd path - alternative):
WSL Users: Use the macOS/Linux config above.
Option 2: Global Install (Most Reliable)
Install globally for offline support and faster startup:
Find your global npm path:
Then add to Claude Desktop config:
Replace YOUR_GLOBAL_PATH with the output from npm root -g.
Option 3: Automatic Installer
For first-time setup, the installer configures Claude Desktop automatically:
After any installation method, restart Claude Desktop completely (quit and reopen).
Related MCP server: ClaudeKeep
Custom Database Location
By default, memories are stored at ~/.memory-mcp/memory.db on all platforms.
To use a custom location, add the env field to your config:
The database is a single portable SQLite file. Back it up by copying the file.
Tools
Tool | Description |
| Store a memory with auto-summarization and entity extraction |
| Search memories with token-aware loading |
| Soft-delete a memory (preserves audit trail) |
Features
FTS5 full-text search (fast, no embeddings needed)
Token budgeting for context-aware responses
Automatic entity extraction and summarization
Soft deletes with provenance tracking
Hybrid relevance scoring (recency + importance + frequency)
Environment Variables
Variable | Default | Description |
| Platform-specific | Database file location |
|
| Default memory expiration |
Troubleshooting
Tools not appearing in Claude Desktop?
Restart Claude Desktop completely (quit and reopen)
Verify config file syntax is valid JSON
Check that Node.js 18+ is installed:
node --version
"Connection closed" on Windows?
Windows requires either:
The
cmd /cwrapper method (see Windows config above), ORThe full path to
npx.cmd(e.g.,C:\Program Files\nodejs\npx.cmd)
Getting stale cached versions?
The npx github: method bypasses npm cache. Alternatively:
Clear npm cache:
npm cache clean --forceUse global install for version control
Slow startup with github: method?
First run requires downloading and installing dependencies (can take 30+ seconds). Subsequent runs are faster but still fetch from GitHub. For faster startup, use the global install method.
Dependencies
@modelcontextprotocol/sdk- MCP protocol implementationbetter-sqlite3- Fast native SQLite with FTS5
Links
Related Projects
momentum - Instant context recovery (<5ms)
AgentForge - Visual agent config builder
Substratia - Memory infrastructure for AI
Support
GitHub Issues - Bug reports & feature requests
GitHub Sponsors - Support development
Disclaimer
This project is provided as-is. It is actively maintained but may have breaking changes between minor versions.
License
MIT