[ docs · ai agents ]
Connect your AI agent (MCP)
Your organization has an MCP server at https://<org>.flow.dexadata.ai/mcp. AI agents connected to it can read your DAGs, runs and logs as you, read-only. Your agent's provider runs the model, so nothing here is billed by Dexaflow.
Claude.ai and ChatGPT
- Copy the server URL from AI agents in the console.
- Claude.ai: Settings → Connectors → Add custom connector (on Team and Enterprise, an owner adds it). ChatGPT: add it as a custom connector in its connector settings, where your plan allows custom connectors.
- Sign in to Dexaflow Cloud and choose Allow. Access is read-only.
- To disconnect, remove the connector in the app. Access also ends when you leave the organization.
Claude Code
claude mcp add --transport http dexaflow-<org> https://<org>.flow.dexadata.ai/mcp
Then run /mcp in Claude Code and choose Authenticate.
Cursor, VS Code, CI and headless use
Create a token in AI agents (30 or 90 days) and use the snippet the page shows. The token is shown once; store it as a secret. Revoke it in the same page.
What the agent can do
List DAGs, diagnose failed runs, search logs. It cannot change, run, pause or deploy anything yet.
Try
- "Which of my pipelines failed in the last 24 hours, and why?"
- "Diagnose the latest failed run of <a DAG>."