mcp-memory
🔥🖥️ MCP Memory is a MCP Server that gives MCP Clients (Cursor, Claude, Windsurf and more) the ability to remember information about users (preferences, behaviors) across conversations.
claude mcp add --transport stdio puliczek-mcp-memory npx -y github:puliczek/mcp-memory
How to use
MCP Memory provides a memory layer for MCP clients by storing user-supplied text and its vector embeddings, then enabling retrieval of relevant memories across conversations. It leverages Cloudflare Workers, Vectorize for vector search, and D1 for persistent storage, orchestrated by a Durable Object and the Agents framework to handle MCP protocol interactions. You can store memories as you chat and retrieve related memories when needed, with similarity-based search ensuring conceptually related items rise to the top even if exact keywords differ.
How to install
Prerequisites:
- Cloudflare account with access to Workers, Vectorize, D1, and Workers AI
- Wrangler CLI installed (via npm: npm install -g wrangler) or using the npm scripts provided by the project
Installation steps:
- Clone the repository or create a new project from this template
- Install dependencies:
npm install
- Create a Vectorize index (dimensions 1024, metric cosine):
npx wrangler vectorize create mcp-memory-vectorize --dimensions 1024 --metric cosine
- Start the development server locally (requires Cloudflare credentials):
npm run dev
- Deploy to Cloudflare once ready:
npm run deploy
Note: If you prefer a one-click setup, use the Cloudflare deployment button in the repository README and follow the prompts to connect your Cloudflare account and select a Vectorize index.
Additional notes
Tips and considerations:
- Ensure your Cloudflare account has access to Vectorize, D1, and Workers AI enabled for optimal performance.
- Memory isolation is per-user, so each user’s memories won’t mix with others.
- The rate limit mentioned (e.g., 100 req/min) is configurable in wrangler.jsonc; adjust as needed for your usage.
- When running locally, some features like vector search or Workers AI may require Cloudflare service connections.
- You can customize authentication and add additional security layers if needed beyond the built-in userId-based protection.
- The MCP protocol integration relies on the Agents framework; make sure your MCP clients are compatible with the expected protocol version.
Related MCP Servers
mcp
MCP server for the Cloudflare API
mcp-stytch-consumer-todo-list
Workers + Stytch TODO App MCP Server
justcall
JustCall's Official MCP Server
mcp-doc-forge
MCP server that provides doc forge capabilities
google -remote
Collection of Google-native tools (e.g., Gmail, Calendar) for the MCP
AI-web mode
一个基于 MCP (Model Context Protocol) 的智能对话助手Web应用,支持实时聊天、工具调用和对话历史管理。