@turbomem/mcp

· Documentation

Give Claude a private, local long-term memory. This is a Model Context Protocol server built on turbomem. Claude can remember durable facts about you and recall them in future conversations. Everything is stored on your own computer - nothing is uploaded anywhere except the AI provider you choose to read and organize your notes.

Ships as a one-click Claude Desktop extension (.mcpb), so no terminal or config files are required.

Install in Claude Desktop (no terminal needed)

1. Download turbomem.mcpb (from the Releases page).
2. Double-click the file, Claude Desktop opens an install screen. (You can also drag it onto Claude Desktop, or go to Settings → Extensions → Advanced settings → Install Extension…)
3. On the install screen, fill in two things and click Install:
   • AI provider - type openai, google, or anthropic (leave as openai if unsure).
   • API key - paste the key for that provider.

That's it. Start a chat and talk normally. Claude will save and recall memories on its own. Try:

"Remember that my dog is named Rex."

then, in a new chat:

"What's my dog's name?"

Which API key do I need?

You only need one key, matching the provider you typed:

| Provider | Where to get a key | What it powers | | ----------- | ------------------------------------------- | ----------------------------------- | | openai | https://platform.openai.com/api-keys | Understanding + searching memories | | google | https://aistudio.google.com/app/apikey | Understanding + searching memories | | anthropic | https://console.anthropic.com/settings/keys | Understanding memories (extraction) |

openai and google use a single key for both understanding and search.

Using Anthropic (Claude)? Anthropic has no embedding model, so search needs a second provider. On the install screen set AI provider = anthropic (+ your Anthropic key), then Search provider = openai or google (+ that provider's key). Extraction runs on Claude and search runs on OpenAI/Google over HTTP — no local model needed.

Your memories live on disk at ~/.turbomem/data by default, the same store used by the turbomem CLI.

What Claude can do

| Tool | What it does | | ------------------- | ----------------------------------------------------- | | remember | Saves durable facts about you to local memory | | recall | Searches your memory for anything relevant to a query | | list_memories | Lists everything currently stored | | forget | Deletes one memory by id | | forget_everything | Erases all memories (only on explicit confirmation) |

Claude is instructed to use these automatically, but you can always ask it directly, e.g. "list everything you remember about me" or "forget that I like dark mode."

Advanced: manual setup

Prefer to wire it up yourself, or using another MCP client? Run the server over stdio.

npm install -g @turbomem/mcp

Add it to your Claude Desktop config (claude_desktop_config.json):

{
  "mcpServers": {
    "turbomem": {
      "command": "turbomem-mcp",
      "env": {
        "TURBOMEM_PROVIDER": "openai",
        "TURBOMEM_API_KEY": "sk-...",
        "TURBOMEM_USER_ID": "me",
        "TURBOMEM_DATA_DIR": "/Users/you/.turbomem/data"
      }
    }
  }
}

Anthropic extraction with OpenAI search (no local model):

{
  "mcpServers": {
    "turbomem": {
      "command": "turbomem-mcp",
      "env": {
        "TURBOMEM_PROVIDER": "anthropic",
        "TURBOMEM_API_KEY": "sk-ant-...",
        "TURBOMEM_EMBEDDINGS_PROVIDER": "openai",
        "TURBOMEM_EMBEDDINGS_API_KEY": "sk-..."
      }
    }
  }
}

Or run without installing:

npx @turbomem/mcp

Prefer Anthropic search fully on-device? Set TURBOMEM_EMBEDDINGS_PROVIDER=local and install the local model:

npm install -g @huggingface/transformers

Environment variables

| Variable | Default | Notes | | ------------------------------ | ------------------ | ------------------------------------------------------- | | TURBOMEM_PROVIDER | openai | Extraction provider: openai, google, or anthropic | | TURBOMEM_API_KEY | - | API key for the extraction provider (required) | | TURBOMEM_EMBEDDINGS_PROVIDER | mirrors provider | Search provider: openai, google, or local | | TURBOMEM_EMBEDDINGS_API_KEY | - | API key for the search provider (needed unless local) | | TURBOMEM_USER_ID | me | Label for this memory profile | | TURBOMEM_DATA_DIR | ~/.turbomem/data | Where memories are stored on disk | | TURBOMEM_EXTRACTION_MODEL | provider default | Override the fact-extraction model | | TURBOMEM_EMBEDDING_MODEL | provider default | Override the embedding model |

Standard provider env vars (OPENAI_API_KEY, GEMINI_API_KEY, ANTHROPIC_API_KEY) are also honored as fallbacks. With openai/google, one key covers both extraction and search; anthropic needs a separate search provider (or local).

Privacy Policy

turbomem is local-first: memories stay on your machine at ~/.turbomem/data by default. We do not receive your memories, chat content, or API keys. Text is sent only to the AI provider you choose (OpenAI, Google, or Anthropic) to extract facts and embed them for search.

• turbomem Privacy Policy
• Documentation copy

Third-party provider policies also apply to API traffic: OpenAI, Google, Anthropic.

You can delete memories with the forget and forget_everything tools, or by removing your data directory.

License

Apache License 2.0