Skip to main content
If your team has its own internal tools or data sources — something not covered by the standard connections — you can plug it in as a custom MCP server.
MCP (Model Context Protocol) is an open standard for safely exposing a tool or data source to an AI coworker. Think of it as a well-defined, secure doorway between your systems and your coworkers.
This is a more technical setup — you, or whoever manages your team’s tools, add it directly in the console. After that, everyone’s coworkers can use it. You can register up to 10 custom servers per account.

​
How it works

1

Add the server address

In the console, enter the address of your MCP-compatible server (a standard HTTP or SSE endpoint), and how it authenticates — no auth, an API key header, or a bearer token.
2

The console verifies it

A real handshake confirms the server is reachable and responds correctly — it isn’t usable by a coworker until this check passes.
3

Your coworkers can use it

Once verified, any coworker on your team can draw on it when relevant.
Credentials are encrypted and never shown again after saving — the console only ever confirms one is set, not what it is. Changing the server address or its credentials resets it to unverified until you test it again.

​
Connection status

This is aimed at teams with their own tools to expose — in practice that often means looping in whoever manages your systems, but nothing about it requires an admin role. If you just want a ready-made connection, see Connections Hub instead.

​
A concrete example

Say your team has an internal pricing database that no standard connection covers. Once it’s exposed as an MCP server and registered here, you could ask Alex to “build a dashboard from our current pricing data” and he pulls it directly — no export, no manual attachment.

​
What this isn’t for

This is for exposing your own systems, not for connecting well-known services — those go through the Connections Hub instead, since it’s a one-click OAuth flow rather than a technical setup. And a coworker only draws on a custom server when the task actually calls for it — connecting one doesn’t change how a coworker behaves otherwise.

​
How this connects

Once verified, a custom server works the same way for every coworker as a Connections Hub service does — see Skills for how a coworker decides what to reach for on a given task.