Connectors

Connecting and importing

Open a connector card, complete the connect flow (API key, token, or OAuth), then choose what to import and which project folder it should land in. Imported files behave exactly like uploads — convert, train, and process them normally. Manual re-sync refreshes content on demand, and disconnecting runs through the backend so it always succeeds.

Output shape and auto-push

Where a connector supports it (TeamSupport, Confluence, and others), choose Combined output — one merged Markdown file, sharded around ~10MB — or Individual files, and toggle auto-push-to-agents so new imports queue for training automatically.

Sync history and cancellation

Every connector records sync history in a shared format: when each import ran, how many items it touched, and any errors. Long-running imports can be cancelled mid-flight, and most connectors resume from where they stopped. Internal admins can tune per-connector pacing from Admin → Rate Limits.

MCP Server connector

Point Command at any remote Model Context Protocol server over Streamable HTTP and ingest its output as Markdown — either by picking resources from resources/list, or by calling tools and saving the responses.

Agent Access (MCP)

Salesforce, Zendesk, Zoho, HubSpot, Jira, Bitbucket, Gmail, tldv, and Zapier expose an Agent Access tab that reuses the existing connection's credentials. Tools are filtered by the scopes the token actually holds, and an optional 'Allow agent to create/update records' toggle gates write tools.

Wiring a connector to an external agent

Option A (recommended): on the connector's page in Command, click Connect to <agent> — a key is minted, the MCP server is registered on the agent, and the binding is tracked for one-click revoke. Option B: create a key from the connector's Agent Access tab and paste the JSON config into the agent's MCP servers list. Revoking the key from the connector page or Settings → Action Keys cuts access immediately.