Membase MCP
Membase keeps one shared memory for AI agents that Claude Code, Cursor and Codex read through MCP, an SDK or the membase command
- 107 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Stores everything written to memory: notes, documents, chat exports; in hosted mode they sit in the Membase account
- The local engine works with an OpenAI key, and a remote connection with a Membase key gets write access
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/membaseDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Stores everything written to memory: notes, documents, chat exports; in hosted mode they sit in the Membase account
- The local engine works with an OpenAI key, and a remote connection with a Membase key gets write access
This is third-party code. Review the repository files before installing.
What it does
Membase is a long-term memory layer for AI agents. Facts, documents, web pages and chat exports from Claude or ChatGPT go into one memory, split into project containers. The repository holds Python and TypeScript clients, the membase command and an MCP server. The memory lives in a Membase account, or on your own machine if you install the package with the local extra. The add, search and ask commands write notes, search them and answer questions over the memory. The media command adds photos, video and audio, in local mode only. In Claude Code the server is connected with claude mcp add: locally through membase --local mcp or remotely over HTTP with a key in the header. Without the header the remote connection asks you to sign in and is read-only. Integrations for LangChain, Dify, Hermes Agent, OpenClaw, Microsoft Agent Framework and Strands Agents sit alongside.
Who it is for. Developers who switch between Claude Code, Cursor and Codex and are tired of explaining the project to each agent again.
Good fit when
- An architecture decision was made in Claude Code and Cursor needs to know about it the next day
- You have a history of Claude or ChatGPT chats and want to turn it into searchable memory
- Notes from different projects get mixed and need splitting into separate containers
- The memory cannot go to the cloud, so local mode with storage in the home folder fits
- Your own Python or TypeScript app should write and read the same memory the agents use
Not a fit when
- You only need rules for one repository: an instructions file next to the code is enough
- Python is older than 3.10, or local mode runs on a version other than 3.12 and 3.13: the packages will not install
- You have no Membase key and no OpenAI key but need memory right now
Example request
Remember in the Engineering container that we picked Postgres for the ledger service, then find that decision from CursorMCP
- Transport
- stdio, http
- Authentication
- key is optional
Environment variables
MEMBASE_API_KEYsecret- Key for hosted Membase memory
OPENAI_API_KEYsecret- Needed by the local memory engine
Limitations
- The hosted mode needs the MEMBASE_API_KEY key; without an authorization header the MCP connection is read-only.
- The local engine installs as unibaseio-membase[local] on Python 3.12 or 3.13 and needs OPENAI_API_KEY; its compiled core ships under a separate Unibase licence.
- The membase package on PyPI belongs to a different project; install unibaseio-membase.
- Media memory works locally only and needs the media extra.
How to disable. claude mcp remove membase disconnects the server in Claude Code. Local memory sits in ~/.membase, hosted memory in the Membase account.
FAQ
How do I connect Membase to Claude Code?
Run claude mcp add --transport http membase https://api.app.membase.com/mcp-http with an Authorization: Bearer header and your key. The local variant connects with claude mcp add membase -- membase --local mcp, after installing the package with the local extra.
How is local Membase memory different from the hosted one?
Local memory lives in ~/.membase and works with commands that start with membase --local. It needs Python 3.12 or 3.13 and an OpenAI key. Hosted memory sits in the Membase account and needs MEMBASE_API_KEY. Photos, video and audio can be added to local memory only.
Which clients read the shared memory?
Python and TypeScript code through the SDK, and Claude Code, Cursor, Codex and ChatGPT through MCP. Setup for each client is described on the Connect your AI page of the documentation.
What are containers in Membase?
Containers split memory by project. Writes and searches accept a container parameter such as Engineering, so notes from one project do not mix with another.
Can I move chat history into Membase?
Yes: the import command accepts chat exports from Claude and ChatGPT as well as Markdown and JSON. It needs local mode with the unibaseio-membase[local] package.
Discussion
No questions or reviews yet.
Related
MCP serversLocal memory for agents: stores conversations and project files verbatim and retrieves them with semantic search over MCP
Shared local memory for AI agents: Claude Code, Codex, Cursor and others find the agreements of past sessions in it and record new ones
Self-hosted long-term memory for agents: semantic search, a knowledge graph and autonomous consolidation over MCP and a REST API
Agent memory as a graph of linked facts: decisions and errors outlive the session, with no vector database or paid API calls
Long-term memory for Claude Code, Cursor and Codex: the agent saves decisions and finds them again in later sessions
Long-term memory for agents: a remote MCP server, plugins for Claude Code, Cursor and Codex, and an API for your own apps
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Membase (this card) | yes | key is optional | 107 | |
| Mem0 | no | 68,514 | plugins and an SDK that carry context between sessions, while Membase stresses one memory for every client and a local mode | |
| Supermemory | yes | OAuth | 31,555 | a remote MCP server, plugins and an API for apps, while Membase also has a fully local engine |
README in short
The README starts with the idea of one memory for every AI and a three-step scheme: add, manage, use. Then come results on the LoCoMo, LongMemEval and DMR benchmarks, a quickstart in Python and TypeScript, command examples, an MCP section and a documentation list. Framework integrations live in the integrations folder.