Give Claude Persistent Memory with Knol MCP Server

Why Claude Needs Persistent Memory

Every time you start a new Claude conversation, you start from scratch. Claude doesn't remember your name, your projects, your coding preferences, or the debugging session you had yesterday. You have to re-explain context every single time.

Knol's MCP server fixes this. Once connected, Claude can store and retrieve memories across sessions — facts, preferences, project context, and relationships. It's like giving Claude a brain that persists.

Setting Up Knol (60 Seconds)

First, get Knol running locally:

git clone https://github.com/aiknol/knol.git
cd knol
docker compose up -d

That's it. Knol is now running on localhost:3000 with PostgreSQL, vector search, knowledge graphs, and the full context engineering stack.

Connecting to Claude Desktop

Open your Claude Desktop MCP configuration:

# macOS
code ~/Library/Application\ Support/Claude/claude_desktop_config.json

# Windows code %APPDATA%/Claude/claude_desktop_config.json ```

Add the Knol MCP server:

{
  "mcpServers": {
    "knol-memory": {
      "command": "npx",
      "args": ["@aiknol/knol-mcp-server"],
      "env": {
        "KNOL_API_URL": "http://localhost:3000",
        "KNOL_API_KEY": "your-api-key"
      }
    }
  }
}

Restart Claude Desktop. You should see the MCP tools icon appear, indicating Knol is connected.

What Claude Can Do Now

With Knol connected, Claude has access to six memory tools:

**knol_store_memory** — Claude can save important facts from your conversation. "User prefers TypeScript over JavaScript" or "Working on a React dashboard for project Atlas."

**knol_search_memory** — Before answering, Claude can search past memories. "What do I know about this user's tech stack?" returns relevant context from previous sessions.

**knol_get_user_context** — Pull a complete summary of everything known about the current user. Preferences, projects, recent interactions, and relationships.

**knol_graph_query** — Traverse the knowledge graph. "What projects is this user connected to?" or "Who else works on project Atlas?"

Real Example

Session 1 (Monday):

You: I'm building a dashboard with Next.js and Tailwind. The backend is Rust with Axum.
Claude: [stores: user tech stack = Next.js, Tailwind, Rust, Axum]

Session 2 (Wednesday):

You: Can you help me add a new API endpoint?
Claude: [searches memory, finds Rust/Axum preference]
Claude: Sure! Since you are using Axum, here is the endpoint...

No re-explanation needed. Claude already knows your stack.

Beyond Claude Desktop

The same MCP server works with Cursor, Windsurf, and any MCP-compatible tool. Your memory is shared across all of them — context from a Claude conversation is available when you are coding in Cursor.

Privacy and Self-Hosting

All memory stays on your machine. Knol runs locally, your data never leaves your infrastructure. For teams, deploy Knol on your own servers with multi-tenant isolation, encryption at rest, and full audit logging.

The MCP server is open-source. Star us on GitHub to follow development.