Installation and configuration
Cloud connection
You do not need Python, uv, a local server, or a BioNodulo administrator secret to connect an AI client to your cloud account. Add this HTTPS MCP URL in your client:
https://bionodulo.com/api/mcpWhen the client prompts you to connect, sign in to BioNodulo in the browser and review the requested permissions. The connection uses your account and selected team; read, write, and run operations require the corresponding authorization. A default login may grant read access only. See Connecting AI clients for client-specific steps and how to request write and run access.
After connecting in ChatGPT or Claude, ask the assistant to open your BioNodulo workbench. If that chat supports MCP Apps, open_bionodulo can display the cloud workbench inside the conversation. Otherwise, use the ordinary cloud tools. This embedded view uses the same account and team authorization as the tools.
If the client cannot connect, confirm that the URL includes /api/mcp, that the remote service is available, and that your client supports remote MCP with browser-based OAuth. If you change teams or permissions, reconnect in the client so it can request the appropriate access.
Local desktop connection
The remote server does not connect to 127.0.0.1 on your computer. Desktop-only tools require a local MCP connector running beside the BioNodulo desktop app. Use the local connector when you need the desktop node catalog, local queue, or local execution state. Its setup is separate from the cloud account connection.
Download the Claude Desktop extension (.mcpb) and install it through Claude Desktop Settings → Extensions. The BioNodulo desktop app must be running on the same computer. This extension only exposes local desktop_* tools; it does not connect to your cloud account.
The BioNodulo source repository includes the extension build script at mcp/scripts/build_native_plugins.py. Developers can also use the mcp/ Python package directly. Keep any backend Clerk secret in server-owned infrastructure. Do not put it in an AI client’s MCP configuration or share it with connector users.