How to use the Memory MCP server for persistent context

Open your MCP host's configuration file and add a memory entry. The server requires no API keys or external services — it stores everything in a local JSON file. The command uses npx with the -y flag for automatic installation. By default, the memory file is stored in the server's working directory.

1{
2  "mcpServers": {
3    "memory": {
4      "command": "npx",
5      "args": [
6        "-y",
7        "@modelcontextprotocol/server-memory"
8      ]
9    }
10  }
11}

Fully quit and reopen Claude Desktop, Cursor, or VS Code. The Memory server provides tools for creating entities, adding relationships, storing observations, and querying the knowledge graph. These should appear in your available tools list after restart.

Tell the AI about your project, team, or preferences. The AI will use the Memory server to create entities and store observations. Be explicit about what you want remembered — the AI does not automatically save everything from the conversation.

1Example prompts:
2
3"Remember that I'm working on a project called ShopFlow, it's an e-commerce platform built with Next.js and Supabase"
4
5"Remember that our team has 3 developers: Alice (frontend lead), Bob (backend), and Carol (DevOps)"
6
7"Remember that I prefer TypeScript strict mode, Tailwind CSS, and shadcn/ui for all my projects"

Close the current conversation and start a new one. Ask the AI about something you stored previously. The AI will query the knowledge graph and recall the information, even though this is a brand new chat session. This demonstrates the persistence of the memory across sessions.

1Example prompts (in a new session):
2
3"What do you know about my ShopFlow project?"
4
5"Who are the members of my development team?"
6
7"What are my coding preferences?"

The knowledge graph supports relationships between entities, creating a connected web of information. Tell the AI how different entities relate to each other — projects to team members, technologies to projects, or any other connections. This helps the AI understand context more deeply.

1Example prompts:
2
3"Remember that Alice is the frontend lead on ShopFlow and she specializes in React performance optimization"
4
5"Remember that ShopFlow depends on Stripe for payments and SendGrid for emails"
6
7"Remember that we decided to use Supabase RLS instead of a custom auth middleware — this was decided on March 15"

Over time, some stored information may become outdated. You can ask the AI to update observations, remove entities, or clear specific relationships. You can also ask it to summarize everything it knows to audit the knowledge graph.

1Example prompts:
2
3"Show me everything you remember about my projects and team"
4
5"Update: we migrated ShopFlow from Supabase to Firebase last week"
6
7"Forget everything about the old ProjectX — we abandoned that project"