Connect your AI assistant
Connect the Claude, ChatGPT, Grok, or Gemini Spark subscription you already pay for to your rundown — no API key, no extra bill.
OnAirFlow does not run AI for you and does not charge for it. Instead, you connect the AI assistant you already pay for — Claude, ChatGPT, Grok, or Gemini Spark — to OnAirFlow's hosted MCP endpoint. Your assistant can then read and edit your rundown directly, in the same chat window you already use. Inference is billed to your own subscription; OnAirFlow never sees a provider credential.
This works on every plan — Creator, Studio, and Newsroom — at no additional cost.
Connect an assistant
Any team member can connect an assistant. Add OnAirFlow as a custom connector in your assistant of choice, using this URL:
https://api.onairflow.com/mcp
The client discovers OnAirFlow's OAuth endpoints, registers itself, and sends you to a consent screen where you approve the connection and the scopes it gets — read-only is an option. There is no API key to copy, nothing to paste, and nothing to renew; tokens rotate automatically. Settings → AI → Assistants in OnAirFlow shows the connector URL and the exact steps for each assistant.
Where you add the connector, and what you need to have, differs by assistant:
| Assistant | Where | Requirements |
|---|---|---|
| Claude | Settings → Connectors → Add custom connector | Free, Pro, Max, Team and Enterprise. Free accounts are limited to one connector. |
| ChatGPT | Settings → Connectors → Advanced → turn on Developer mode, then add as an MCP server | Pro, Plus, Business, Enterprise and Edu. |
| Grok | grok.com/connectors → Add a custom MCP connector | Paid tiers, on web, iOS and Android. |
| Gemini Spark | Settings → Connected apps → Add a custom app | US only, 18+, and personal Google accounts only — Workspace accounts cannot add custom apps. |
Each connection appears as its own named agent under Settings → AI → Assistants, and is revocable there at any time.
What your assistant can do
A connected assistant can never do more in OnAirFlow than you can. A host's assistant is bounded by that host's access; a producer's by the producer's. An admin's assistant is capped at producer level — admin powers are deliberately not delegated to an assistant. Revoking a person's access revokes their assistant's access with it.
Within those bounds, your assistant works on the rundown surface: shows, stories and their notes. Managing columns and uploading attachments require producer or admin access, so a host's assistant cannot do them. In practice that means asking for things in plain language:
- "What's on tomorrow's show?"
- "Summarize these three source links into four on-air bullets and put them in the host notes for the lead story."
- "File these stories into Ideas and move the weather block up."
Writing host notes from source links is one of the most common uses — paste the links into your assistant, ask for bullets in the voice you want, and have it write the result straight into the story's Notes field.
Why it works this way
Consumer AI subscriptions are licensed to the person who buys them. As we read the major providers' terms, a hosted application like ours may not route its own requests through your subscription, or collect, store, or intermediate the credentials behind it. Connecting your assistant to OnAirFlow is the direction that works: the request originates from your own assistant, on your own plan, and OnAirFlow is simply a tool it can reach.
The practical upshot is simpler. You get the frontier model you already chose, at the price you already pay, with nothing extra to meter on our side. Your assistant's own usage limits still apply.
Running your own agents
If you are wiring up automations rather than chatting — Claude Code, Cursor, or a fleet of your own agents — use a scoped API key instead. Admins mint them under Settings → AI → Agents & keys, where each agent gets its own key (shown once) and scopes of rundown:read and/or rundown:write.
For the full setup — REST API, hosted MCP, connectors, and the ready-made operator skill — see the agent documentation.
Troubleshooting
- The consent screen rejects the connection — you can have at most 10 connected assistants at once. Revoke one under Settings → AI → Assistants and try again. A consent link that was left sitting can also expire; start the connection again from your assistant.
- Your assistant can read but not write — the connection was approved with read-only scope. Revoke it under Settings → AI → Assistants and reconnect, granting write access.
- A connection stopped working — someone may have revoked it, or your own permissions may have changed. Reconnect from Settings → AI → Assistants.
Related
- Notes your assistant writes live in the story's Notes field, which exports with the rundown — see Exports & archive.
- Notes are searchable along with titles — see Search & notifications.
- Assistant permissions follow team roles — see Team roles & permissions.