Using GATE
Connecting outside tools (MCP)
MCP lets you give your bot extra abilities borrowed from an outside service. Connect one, grant it to a bot, and from then on that bot can reach for the connected tool in every conversation — not just answer questions about it, but actually act.
What connecting a tool does
Once a tool is connected and granted to a bot, it shows up as something the bot can use in the background whenever it's relevant — in chat, in coding sessions, anywhere that bot works. You don't invoke it directly; you just ask for what you want, and your bot reaches for the tool if it needs it.
Ask: "Check what's in our Linear backlog and start on the top bug," and your bot uses the connected Linear tool itself, without you naming it.
The catalog
A handful of popular services are one click away:
- Webflow, Notion, Linear, GitHub, Stripe — paste a token from that service and you're connected.
- Figma — no token to paste; click Authorize and complete the sign-in in your browser instead.
If what you need isn't in the catalog, a custom server option accepts a free-form configuration for anything else that speaks MCP.
Granting a tool to a bot
Connections and grants both happen from the bot's page in the dashboard. Connecting makes a tool available in your organization; granting is what actually turns it on for a specific bot. No grant, no access — a connected tool a bot hasn't been granted stays invisible to it.
You can grant the same connection to more than one bot, and revoke it from any of them at any time.
Reviewing what your bot did
Every time a connected tool is actually used, it's recorded in an audit log you can review from the connection or from the bot's page — so if your bot books something in Notion or opens a GitHub issue, there's a trail of exactly what it called and when.
These credentials do reach the bot
For GATE's regular integrations (Google, Slack, Stripe billing, and the rest), your credentials never touch the bot — GATE holds them and acts on the bot's behalf. MCP is the exception. An MCP connection's token lives in that bot's own environment, because the outside tool's server process runs alongside the bot and needs the real credential to talk to the service.
Only connect tokens you'd be comfortable handing directly to that bot, and prefer scoped, limited-permission tokens over broad ones wherever the service allows it.
See Credentials & delegation for how the rest of GATE's integrations keep credentials off the bot entirely — MCP is the one place that pattern doesn't hold.
Going deeper
- Built-in tools — the rest of what your bot can do on request
- Credentials & delegation — the broader credential model, and why MCP is the exception to it