Skip to content
MCP server

MCP Memory

Local SQLite memory for AI agents with BM25 search and no external services

Runs locallyStatus and history
Install
Pricing
Open source
Platforms
CLI, self-hosted, Linux
Runs
Locally
Version
0.6.0
Listing updated
today

Similar MCP servers

About MCP Memory

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

MCP Memory keeps an agent's notes in a single SQLite file on your own machine, searched with FTS5 and BM25 ranking. The author's pitch is that agents forget everything between conversations and existing fixes are cloud-based or tied to one client, so this one is deliberately small: plain text, tags and free metadata, grouped into namespaces per project or agent.

Nine tools cover saving, searching, updating, deleting, listing, recent items, stats, and JSONL export and import. Tools carry MCP read-only and destructive hints so clients can auto-approve reads and ask before edits. It runs over stdio or as a local streamable-HTTP server on port 8765, and several clients can share the same database file. The registry also lists a container image.

The README is frank about limits. Search is lexical, so synonyms with no shared words will not match. There is no semantic search, no images or PDFs, no multi-tenancy, no sync between machines, no UI and no authentication, so exposing it to the internet needs a proxy. It positions itself against the official server-memory knowledge graph as a text-memory alternative.

Install

Run in a terminal

claude mcp add mcp-memory -- docker run -i --rm ghcr.io/germaniu/mcp-memory:0.6.0

Add to claude_desktop_config.json

{
  "mcpServers": {
    "mcp-memory": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "ghcr.io/germaniu/mcp-memory:0.6.0"
      ]
    }
  }
}
Install in CursorOpens Cursor, which asks you to confirm. Or add it by hand:

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "mcp-memory": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "ghcr.io/germaniu/mcp-memory:0.6.0"
      ]
    }
  }
}
Install in VS CodeOpens VS Code, which asks you to confirm. Or add it by hand:

Add to .vscode/mcp.json

{
  "servers": {
    "mcp-memory": {
      "type": "stdio",
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "ghcr.io/germaniu/mcp-memory:0.6.0"
      ]
    }
  }
}

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

{
  "mcpServers": {
    "mcp-memory": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "ghcr.io/germaniu/mcp-memory:0.6.0"
      ]
    }
  }
}

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

{
  "mcpServers": {
    "mcp-memory": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "ghcr.io/germaniu/mcp-memory:0.6.0"
      ]
    }
  }
}

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

Configuration

  • DB_PATHPath to the SQLite file inside the container. Keep it under the mounted volume (/data) so memories survive restarts.

Tools this server provides (9)

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

memory_save
Save text with tags and metadata
memory_search
BM25 lexical search with namespace filter
memory_update
Change content, tags or metadata by id
memory_delete
Delete a memory by id
memory_list
List memories by namespace, paginated
memory_recent
Show the latest memories by update time
memory_stats
Counts, namespaces and oldest and newest entries
memory_export
Export memories as JSONL
memory_import
Import a JSONL export, skipping id collisions

What it does

Local SQLite store
One database file holds every memory, with backup by copying it.
BM25 search
FTS5 ranking with namespace filtering.
Namespaces and tags
Separate memory pools per project, with tags and metadata.
Export and import
Move memories as JSONL between machines.
Tool annotations
Read and destructive hints let clients gate approvals.
Two transports
Use stdio or a local HTTP endpoint.

Good for

  • Let a coding agent remember project decisions across sessions
  • Share one memory between Claude Code and Cursor
  • Keep separate memory per project with namespaces
  • Back up an agent's memory by copying a file

Pricing

Open source

Open source, runs locally

MCP Memory FAQ

Does it use embeddings?

No. Search is lexical with BM25, so paraphrases without shared words will not match.

Where is data stored?

In a local SQLite file set by DB_PATH, with no cloud service.

Does it have authentication?

No. HTTP mode listens on localhost, so use a proxy with auth if you expose it.

How many tools does it expose?

Nine, from memory_save to memory_import.

Is MCP Memory free?

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

How do I add MCP Memory to Claude Code?

Run this command in your terminal: claude mcp add mcp-memory -- docker run -i --rm ghcr.io/germaniu/mcp-memory:0.6.0

What are the best alternatives to MCP Memory?

The closest alternatives to MCP 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