Cursor Brain

An MCP (Model Context Protocol) server that gives Cursor IDE a persistent memory layer: store conversations and coding decisions, then have the AI retrieve relevant context automatically via hybrid (semantic + lexical) search. Works locally - no cloud required.

How it works

1. You add Cursor Brain as an MCP server in Cursor. Cursor starts the cursor-brain process and talks to it over stdio.
2. The AI gets four tools: memory_search, memory_add, memory_delete, memory_stats. When you say things like "remember that we use Postgres" or "what did we decide about auth?", the model can call these tools.
3. Memory is stored locally in a SQLite database (default: ~/.cursor-brain/storage/memory.db). Each entry has content, optional tags, and an optional embedding. Search uses:
   • Lexical search (FTS5) for keyword match - always works.
   • Semantic search (vector similarity) when you add an API key in ~/.cursor-brain/config.json - improves relevance for natural-language queries.
4. No env vars required. Storage path and optional OpenAI key can be set in ~/.cursor-brain/config.json; otherwise defaults are used and only lexical search runs.
5. Add a Cursor rule (e.g. in .cursor/rules/) so the AI is instructed to call memory_search when answering and memory_add when you ask to remember something. Then the agent uses Cursor Brain by default.

# Integrate cursor-brain

Whenever generating answers or code:

- Before answering, call the MCP tool `memory_search` with the user's query (so the agent retrieves relevant memories).
- When the user asks to remember something, call the MCP tool `memory_add` with that content.
- Use `memory_delete` when the user requests forgetting something or cleaning up.
- Use `memory_stats` to gather internal memory metrics when helpful.

Install

npm install -g @samhithgardas/cursor-brain

Or run without installing:

npx @samhithgardas/cursor-brain

Configure Cursor

Add the MCP server so the AI can use the tools.

Settings > Tools & MCP (or MCP): add a new server with Command: cursor-brain (or Command: npx, Args: ["-y", "@samhithgardas/cursor-brain"]).

Or edit ~/.cursor/mcp.json (user) or .cursor/mcp.json (project):

{
  "mcpServers": {
    "cursor-brain": {
      "command": "npx",
      "args": ["-y", "@samhithgardas/cursor-brain"]
    }
  }
}

Restart Cursor after changing MCP config.

Optional config

To use a custom storage path or enable semantic search (embeddings), create ~/.cursor-brain/config.json:

{
  "storagePath": "/path/to/storage",
  "openaiApiKey": "sk-..."
}

• storagePath: Where memory.db lives. Omit to use ~/.cursor-brain/storage.
• openaiApiKey: Optional. If set, enables vector search; if omitted, only keyword (FTS) search is used.

MCP tools

| Tool | Description | | ----------------- | --------------------------------------------------------------------------------- | | memory_search | Hybrid search; returns relevant memories for a query. | | memory_add | Store a memory (type: session_memory, long_term_memory, or project_memory). | | memory_delete | Delete by id or ids. | | memory_stats | Return total count and counts by type. |

Make the AI use it by default

Add a Cursor rule (e.g. .cursor/rules/cursor-brain.mdc) with alwaysApply: true that tells the agent to:

• Call memory_search with the user's question when answering, to pull in relevant past context.
• Call memory_add when the user asks to remember something or when recording an important decision.

See docs/usage.md for example prompts and flows.

From source

git clone https://github.com/samhith123/cursor-brain.git
cd cursor-brain
npm install
npm run build

Run the server with npm run mcp:run or cursor-brain, and point Cursor's MCP config at the cursor-brain command (or at node /path/to/cursor-brain/dist/mcp/server.js).

Changelog

0.1.6

• Breaking: Tool names changed from dot notation (memory.search) to underscore notation (memory_search) for better MCP compatibility
• Updated documentation to reflect new tool names

0.1.5

• Added embedding dimension tracking
• Improved local embeddings support

Docs

• docs/setup.md - Setup and config.
• docs/usage.md - Example usage and flows.

License

MIT. See LICENSE. Copyright (c) 2026 Samhith Gardas.