Skip to content
MCP server

Local Memory

Local-first persistent memory for Claude, Cursor and Codex, stored in one SQLite file

Runs locallyStatus and history
Install
Pricing
Open source
Platforms
CLI, self-hosted, macOS
Runs
Locally
Version
2.4.4
Listing updated
today

Similar MCP servers

About Local Memory

Local Memory is an open-source MCP server for knowledge and memory. It runs locally on your machine and is started by your AI client.

The pitch is simple: your assistant stops forgetting between chats, and nothing leaves your machine. Everything lives in a single SQLite file whose path you can change with MEMORY_DB_PATH, with a knowledge graph of entities and observations alongside learnings and decisions.

Search is hybrid. FTS5 keyword ranking is fused with vector similarity using reciprocal rank fusion, and the embeddings come from a multilingual-e5-small model that runs on CPU through Transformers.js and is cached at roughly 30 MB. If the vector extension cannot load, or you set MEMORY_EMBED_DISABLED, it drops to keyword-only search. A duplicate gatekeeper prevents storing the same fact twice.

Version 2.1 added time-aware features: an asOf query returns what was known on a given date, and facts can be superseded instead of overwritten. The README is explicit that this is valid-time only, not full bi-temporal. It reports 25 tools and distinguishes itself from an unrelated project with a similar name by its scoped npm name @studiomeyer/local-memory-mcp.

Install

Run in a terminal

claude mcp add local-memory-mcp -- npx -y @studiomeyer/local-memory-mcp

Add to claude_desktop_config.json

{
  "mcpServers": {
    "local-memory-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@studiomeyer/local-memory-mcp"
      ]
    }
  }
}
Install in CursorOpens Cursor, which asks you to confirm. Or add it by hand:

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "local-memory-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@studiomeyer/local-memory-mcp"
      ]
    }
  }
}
Install in VS CodeOpens VS Code, which asks you to confirm. Or add it by hand:

Add to .vscode/mcp.json

{
  "servers": {
    "local-memory-mcp": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "@studiomeyer/local-memory-mcp"
      ]
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "local-memory-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@studiomeyer/local-memory-mcp"
      ]
    }
  }
}

Add to your client's MCP settings (most use this shape)

{
  "mcpServers": {
    "local-memory-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@studiomeyer/local-memory-mcp"
      ]
    }
  }
}

Using several servers? Build one config file with the MCP config generator.

Configuration

  • MEMORY_DB_PATHCustom path for the SQLite database file
  • MEMORY_EMBED_DISABLEDDisable the local embedding pipeline (forces FTS5-only search). Useful in restricted networks where the model can't be downloaded.
  • MEMORY_EMBED_MODELOverride the embedding model id (defaults to Xenova/multilingual-e5-small).
  • MEMORY_EMBED_CACHE_DIROverride the Transformers.js cache directory (defaults to the OS data dir).

Tools this server provides (6)

Taken from the project documentation; the running server may differ.

memory_session_start
Loads context from earlier sessions at the start of a conversation.
memory_session_end
Closes out the session when the conversation is done.
memory_learn
Stores patterns, insights and mistakes.
memory_entity_observe
Records facts about people, projects and tools in the knowledge graph.
memory_search
Hybrid search across stored memory with hybrid, fts or vector modes.
memory_entity_open
Opens an entity, optionally as of a given date.

What it does

Single-file storage
All memory sits in one local SQLite database; no cloud or API keys are involved.
Hybrid retrieval
Combines FTS5 BM25 keyword ranking with local vector similarity via reciprocal rank fusion.
Knowledge graph
Records entities and observations about people, projects and tools.
As-of queries
Opens an entity as it stood on a chosen date using validity windows.
Duplicate detection
Blocks storing the same information twice.
Portable export
Memory can be exported and imported as JSON.
Multilingual embeddings
The default model covers more than 100 languages and runs on CPU.

Good for

  • Carry project decisions across Claude Code sessions
  • Keep notes about people and projects in a private local graph
  • Search past learnings in a different language than they were written
  • Restore what an assistant knew on a past date
  • Share one memory store between Claude, Cursor and Codex

Pricing

Open source

Free; runs locally via npm

Local Memory FAQ

Does it need a cloud account or API key?

No. It runs locally and the README states no cloud and no API keys.

What if the embedding model cannot download?

Set MEMORY_EMBED_DISABLED to force FTS5-only search, which suits restricted networks.

How do I make a session start automatically?

Add a line to CLAUDE.md telling the assistant to call memory_session_start, or use a SessionStart hook.

Which clients are documented?

Claude Code, Claude Desktop, Cursor, VS Code and Codex.

Is Local Memory free?

Yes. Local Memory is open source, so you can use and self-host it for free.

How do I add Local Memory to Claude Code?

Run this command in your terminal: claude mcp add local-memory-mcp -- npx -y @studiomeyer/local-memory-mcp

What are the best alternatives to Local Memory?

The closest alternatives to Local Memory are Codebase Memory, XERJ, Claude Code Ultimate Guide, Nautilus Compass Memory and NeuroStack.

Where this comes from

  • Description: written from the project's README.
  • Install steps: from the official MCP Registry entry.

How AIPoweredBot checks servers