membase — Claude Code MCP server: shared memory for AI
The membase MCP server gives agents one long-term memory: search, profile, adding facts and documents. Handy when you switch between models and tools.
claude mcp add --transport http membase https://api.app.membase.com/mcp-httpWhat membase does
membase keeps one memory that all your AIs read from. You add facts, documents, pages and chat exports, then search, ask questions and drop what should go. Containers keep projects apart.
The MCP server runs two ways. Local starts with membase --local mcp and needs unibaseio-membase[local] on Python 3.12 or 3.13 plus an OPENAI_API_KEY. Hosted connects to https://api.app.membase.com/mcp-http; without the Authorization header the connection is read-only. The agent, import and serve commands need the local package.
Hosted mode needs MEMBASE_API_KEY. The Python package is unibaseio-membase, while membase on PyPI is a different project, so don't mix them up at install time. The license is proprietary.
What it helps with
- stop re-explaining project context when you switch models
- find which database a service uses
- pull a Claude or ChatGPT chat export into one memory
- keep memory separated per project with containers
- ask a question over accumulated documents and get an answer
Example prompts for the agent
- Connect the membase MCP server and look up which database we picked for the ledger
- Add a fact to the Engineering container: we picked Postgres for the ledger
- Import the chat export in
claude-export.jsoninto memory
How to install membase
Platform CLI: remote server
claude mcp add --transport http membase https://api.app.membase.com/mcp-httpPlatform CLI: remote server
codex mcp add membase --url https://api.app.membase.com/mcp-httpConfig file
{
"mcpServers": {
"membase": {
"url": "https://api.app.membase.com/mcp-http"
}
}
}Config file
{
"servers": {
"membase": {
"type": "http",
"url": "https://api.app.membase.com/mcp-http"
}
}
}Platform CLI: remote server
gemini mcp add -t http membase https://api.app.membase.com/mcp-httpStep by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about membase
How do I install membase in Claude Code?
Run claude mcp add --transport http membase https://api.app.membase.com/mcp-http. Without the Authorization header the connection is read-only.
What does membase do?
It keeps one long-term memory for you and your AIs: searching memories, reading the profile, adding facts and documents. It is read through the SDK or MCP.
How does local mode differ from hosted?
Local starts with membase --local mcp and needs unibaseio-membase[local], Python 3.12 or 3.13 and OPENAI_API_KEY. Hosted works over https://api.app.membase.com/mcp-http and needs MEMBASE_API_KEY.
