ScalebrowserDOCS

Which clients work

The MCP clients that can drive a Scalebrowser profile: the ones that run on your machine and reach the daemon directly, and the ones in somebody else's cloud that need a relay.

Everything on this page speaks the same MCP server. What differs is only where the client runs, and that decides which of two paths it takes.

Two paths

Where the client runsHow it reaches the daemon
On your machineStraight to 127.0.0.1, over HTTP with your token, or over stdio.
In somebody else's cloudThrough the relay, over OAuth. See Remote access.

A client in a browser tab cannot reach your loopback address, and no setting inside it changes that. That is the whole distinction.

Clients that run on your machine

Each has a page, because each spells its configuration differently and one of them fails silently if you get it wrong.

Clients in someone else's cloud

These need Remote access, which is off until you switch it on.

ServiceWorksVerified
Claude on the web, desktop, mobileyesDriven end to end, tool call answered
GrokyesDriven end to end
ChatGPTBusiness, Enterprise and Edu onlyDriven on Plus: the switch does not exist there
Gemini Enterprise, Gemini CLIyesVendor documentation
Gemini consumer appnoTakes only partnered connectors
Mistral, now "Vibe"noDriven: the connect button has no effect

Anything else

The daemon speaks the standard Streamable HTTP transport at /v1/mcp and negotiates the protocol revision, so a client not listed here works if it speaks MCP.

What it needs is the address, the bearer token in an Authorization header, and, if you reach the daemon under a public name, that name in [mcp] allowed_hosts.

Next