Skip to main content
Glama

Memory MCP

This repository implements a small MCP server that exposes a mem_save tool for recording structured memory entries to a per-project JSON file (.memory/memory.json). The server is intentionally simple: it stores entries provided by the AI and enforces a token-budgeted FIFO eviction policy.

Installation

npm install npm run build

Configuration

Add this server to your MCP client configuration. Example:

{ "mcpServers": { "memory-mcp": { "command": "node", "args": ["/path/to/mcp/dist/server.js"] } } }

Features

This server does not summarize or decide what to store by itself. User should decide when to call mem_save.

  • Tool: mem_save — record memories for a project (JSON storage)

  • Token estimation and FIFO eviction to keep estimated tokens ≤ 1000

  • Auto-creates .memory/memory.json if missing

mem_save behavior

Inputs:

  • projectPath (string): absolute project path provided by the AI

  • entries (array): list of memory entry objects. Each entry has:

    • what (string)

    • why (string)

    • outcome (string)

    • task_context, constraints, dependencies (optional strings)

Storage:

  • Memory is stored at <projectPath>/.memory/memory.json:

    { "entries": [ ... ], "meta": { "total_entries": number, "estimated_tokens": number, "last_updated": "YYYY-MM-DD" } }

Token estimation (approximate):

  • Chinese chars ≈ 1.3 tokens each

  • English letters ≈ 0.3 tokens each

  • Other symbols ≈ 0.6 tokens each

Eviction policy:

  • After appending new entries, the server computes an estimated total token usage.

  • If the total exceeds 1000 tokens, it removes the oldest entries (FIFO) one-by-one until the total is ≤ 1000.

Return value:

  • On success the tool returns a JSON text message with success: true and a short summary.

License

MIT

Deploy Server
A
security – no known vulnerabilities
A
license - permissive license
A
quality - confirmed to work

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/Artin0123/workflow-mcp'

If you have feedback or need assistance with the MCP directory API, please join our Discord server