Memory MCP
Official persistent memory server for AI — store facts, preferences, and context across conversations using a knowledge graph.
// Add to your client
{
"mcpServers": {
"memory": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-memory"]
}
}
}Paste into your client's MCP configuration file.
// Try it now
"Remember that I prefer pnpm over npm, and I use Biome instead of Prettier for formatting"
- ! Memory files live on disk — protect them the same way you protect any other local data
- ! Conflicting memories accumulate — occasionally review and prune
- ! If you share your MCP config, others may see or overwrite your memories
API key or simple config
// When to use
- You're tired of repeating the same context every session
- You want your AI to remember project-specific preferences
- You're building long-running agents that need state across sessions
- You want personalized interactions over time
// When NOT to use
- You want every conversation to start fresh with no memory
- You work on highly sensitive data that shouldn't persist across sessions
- You share the MCP setup with others — their memories will mix with yours
// Usage Scenarios
Project Preferences
Tell your AI your stack once and have it remember.
Example prompt
"Remember that on this project we use TypeScript, Next.js 15, and tRPC"
People & Relationships
Keep notes about collaborators so context is preserved.
Example prompt
"Remember that Alex is the backend lead and prefers async written updates over meetings"
Ongoing Task State
Track where you left off on multi-session projects.
Example prompt
"Remember that I finished the auth rewrite last week and am now starting on the billing module"
// About
Plain English
By default, an AI assistant forgets you the moment a conversation ends. Tell Claude on Monday that you prefer pnpm over npm, and on Tuesday you have to tell it again. Re-explain the project, the people, the conventions — every single session. This server gives your AI a notebook. When you say something worth keeping — 'we use Postgres, not MySQL,' 'the design lead is Sarah,' 'this project ships every other Thursday' — Claude writes it down in a small local memory file and reads it back into future conversations automatically. Over time, the assistant gets to know your project the way a colleague would. For solo builders, this is the difference between an AI that feels like a stranger every morning and one that feels like a teammate. For ongoing projects, it's a way to teach Claude conventions once and have them stick. The memories live in plain files on your computer, so you stay in control: you can read them, edit them, or wipe them. The trade-off: it works best when you actively curate it. Memory that nobody prunes turns into noise.
The official Memory MCP server lets AI assistants maintain persistent knowledge across conversations using an in-process knowledge graph. Store entities, relations, and observations — and your AI will remember them session to session.
Perfect for building AI assistants that get smarter over time by retaining user preferences, project context, and learned facts.
// Use Cases
- Give Claude persistent memory about your preferences
- Build AI assistants that remember past conversations
- Store project context that persists between sessions
- Create a personal AI knowledge base
// Works With
// Also Consider
// Related Servers
Anthropic Claude API MCP
Use Claude API within Claude — chain AI calls, compare model outputs, and build sophisticated multi-agent workflows.
Sequential Thinking MCP
Official structured thinking server — enable AI to use step-by-step reasoning for complex problem solving and analysis.
OpenAI MCP
Official OpenAI MCP server — access GPT models, image generation, embeddings, and more from any MCP-compatible AI client.