Connect your tools
Give the agent access to external services — from the integrations catalog, or as a connector you add yourself.
Out of the box the agent works with what is on the machine: files, a shell, git. Connecting your tools lets it reach the services the work actually lives in — issue trackers, chat, databases, APIs. Everything on this page lives under Customize → Connect on your machine.
Integrations: the catalog
Integrations is the curated way in. Browse the catalog, pick a service, fill in whatever credential it asks for, and the connection is installed on your machine — its tools appear to the agent immediately. This covers the common cases with no configuration beyond the credential.
Connectors: the advanced way
Connectors is the generic mechanism underneath: MCP servers that give the agent tools reaching outside the machine. Add one yourself when the catalog does not cover a service.
- Local — runs a command on your machine, such as an MCP server installed from npm. You provide the command and any environment variables it needs, one per line.
- Remote — connects to a hosted MCP endpoint. You provide the URL and any headers, one per line.
Each connector shows its status — Connected, Failed, Disabled — and how many tools it contributes. You can enable or disable one without removing it, and manage its tools individually, which matters when a server exposes twenty tools and the agent should see three. Removing a connector deletes it from the machine config, and its tools disappear from the agent immediately.
Secrets
Credentials that several tools need — or that you would rather not paste into a connector definition — belong in Customize → Secrets. A secret is set once on the machine and available to the agent environment; the agent can use it without ever printing it.
Where credentials live
Everything you connect is stored on your machine, not in the platform: connector definitions, environment variables, secrets. Deleting a machine takes its credentials with it.