Because the tool set is every action across all your connected integrations, static mode can be large. Dynamic mode inverts that — the server returns only enable-tools, and the agent enables the specific actions relevant to the query. Scoping by the apps parameter is the simpler lever when you already know you only want, say, Google Calendar and Docs.
The Membrane MCP server: it turns the actions of integrations connected on Membrane into tools an agent can call, over HTTP or SSE.
- Tools for the actions of every integration connected under your Membrane token
- Static mode returns all available tools; dynamic mode (?mode=dynamic) returns one tool, enable-tools, to switch on just the ones a session needs
- Scoping to specific integrations with the apps query parameter, e.g. ?apps=google-calendar,google-docs
- Streamable HTTP at /mcp is the recommended transport; SSE at /sse is supported but deprecated
- Experimental persistent chat sessions via an x-chat-id header
Node.js v18 or higher and a Membrane account. Provide a Membrane access token by query parameter or the Authorization header. Clone, npm install and npm run build to run locally at http://localhost:3000, or deploy the included Dockerfile. Which tools appear depends on which integrations are connected under the token.
One command plus a key — npx -y mcp-remote https://<HOSTED_MCP_SERVER_URL>/sse?token={ACCESS_TOKEN}, then supply credentials