Skip to main content
gbrain
Sign in

The same tools, in any assistant.

Claude Code, Cursor and Codex speak MCP. Point one at this endpoint and it reaches the same accounts, under the same permission and the same daily ceiling, writing to the same record. 

claude mcp add gbrain https://gbrain.io/mcp --header "Authorization: Bearer <token>"

The token comes from the client's own page in the workspace, where it is minted once and shown once.

An assistant sees what the permission allows

The tool list an assistant is handed is built from the permission that client is hooked up to. A tool nobody granted is not described, not offered and not callable.

Refusals come back as structured data rather than prose, so an assistant reads what was missing and asks for it instead of guessing at a retry.

Calls land on the same ledger as the terminal's, against the same daily ceiling, on the client's own page in the workspace.

What it reaches

Prefer a terminal? The command line reaches all of it too, and gbrain Tools is how the two fit together.