# unibaseio/membase-mcp [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/unibaseio/membase-mcp  
**GitHub Stars:** 142  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/unibaseio-membase-mcp

## Description
Save and query your agent memory in distributed way by Membase

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "membase-mcp": {
    "command": "uvx",
    "args": ["unibaseio-membase"],
    "env": {
      "MEMBASE_API_KEY": "",
      "OPENAI_API_KEY": ""
    }
  }
}
```

**Requires environment variables:** `MEMBASE_API_KEY`, `OPENAI_API_KEY` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What unibaseio/membase-mcp does

unibaseio/membase-mcp gives MCP-compatible AI clients access to Membase memory. Agents can save facts, documents, and other knowledge, then search that memory or ask questions about it later. The same account-level memory can be used from the SDK, command line, web app, browser extension, and MCP clients.

Membase supports two storage modes. Hosted usage keeps memory in a Membase account and is accessed through the hosted API. Local usage stores data on the machine, under `~/.membase` by default, and can run through the local engine. Containers provide a way to separate memories for different projects.

## How it works

The MCP connection can point to a local server started with the `membase` command, or to Membase's hosted HTTP endpoint. A local setup uses the `membase --local mcp` command. Hosted access uses `https://api.app.membase.com/mcp-http` and can authenticate with a bearer token. A hosted connection without that header signs the user in and provides read-only access.

The underlying Membase package also provides Python and TypeScript clients and a command-line interface. The Python package requires Python 3.10 or newer for hosted use and Python 3.12 or 3.13 for the local extra. The TypeScript package requires Node.js 18 or newer. Local operation uses an OpenAI API key, while hosted operation uses a Membase API key.

## Setup and configuration

Install the Python package with:

```bash
pip install unibaseio-membase
```

For the local engine, install the local extra instead:

```bash
pip install 'unibaseio-membase[local]'
```

The TypeScript package and standalone CLI are available with npm:

```bash
npm install @unibaseio/membase
npm install -g @unibaseio/membase-cli
```

Add a local MCP connection with:

```bash
claude mcp add membase -- membase --local mcp
```

For hosted MCP access, configure the HTTP endpoint and pass `MEMBASE_API_KEY` as a bearer token. The local installation requires `OPENAI_API_KEY`. The optional `media` extra adds local processing for photos, videos, and audio.

## Tools and capabilities

- Save individual memories and longer documents.
- Search stored memory with natural-language queries.
- Ask questions that use the stored memory as context.
- Organize information into project-specific containers.
- Import Claude, ChatGPT, Markdown, or JSON chat exports through the CLI.
- Ingest and search agent traces with agent-specific commands.
- Process local media for shots, speech, speakers, and semantic search when the media extra is installed.
- Share memory across supported SDK, CLI, web, and MCP access paths.

## Limitations and notes

Hosted and local deployments have different requirements and behavior. Hosted access needs a Membase account and API key for writable authenticated use; a hosted connection without the authorization header is read-only. Local mode requires the local package extra and an OpenAI API key.

Media processing is local-only and requires the separate `media` extra. The Python distribution is named `unibaseio-membase`, while the unrelated PyPI package named `membase` should not be substituted. The main package is MIT licensed; the compiled local engine installed by the local extra has its own Unibase license.

_Full upstream README: https://allmcps.com/mcp/unibaseio-membase-mcp/readme_