Skip to content
Typestream
Try it free

Connected agents

Connected agents

Use Typestream from your own AI agent over MCP: what it can do, turning it on, confirmations, credits, the audit log and revoking access.

Connected agents let you use Typestream from your own AI agent: Claude, ChatGPT, Cursor, or any client that speaks the Model Context Protocol (MCP). Ask it about your coverage and it answers with the articles behind every number, as Typestream's own assistant does.

Your agent gets the same tools Typestream's assistant uses. It acts as you, in one of your organisations, and spends that organisation's credits under the same limits.

Turn it on

Connected agents are off until an owner or admin turns them on in Settings → Connected agents. Turning them off stops every connected client at its next call; their connections stay, so turning it back on needs no reconnecting.

Connect a client

Every client needs the same thing, the server URL:

https://api.dev.typestream.nz/mcp

It's also on the settings page, with a copy button. Add it to your client, sign in to Typestream when it asks, and pick the organisation it should work in. Step-by-step guides:

What it can do

Your agent can find your subjects and edit their profiles, search and read their coverage, research and set up new subjects, backfill them, compile and run questions over coverage (tone, share, counts, over time or by outlet), regroup and rank the answers, check your credits, and create, change, pause and list monitors.

Two tools stay in Typestream's own chat, because they draw on its screen: make_chart (a chart block your client can't show; the numbers come back from aggregate instead) and pin_block (pinning such a block to a dashboard).

It asks before it spends

Anything that would cost more than your organisation's auto-approve limit, and every new or changed monitor, doesn't run on the first call. Your agent gets the price and a one-time confirmation instead, and should ask you. Only when you agree does it call again to go ahead. The confirmation lasts 15 minutes and works only for that exact request.

The limit also counts what an agent has already spent without asking: once one has spent the limit in a day, its next spending call asks too, whatever it costs.

Confirming is up to the client's model, which should always ask you first. Keep the auto-approve limit where you're comfortable, in Settings → Billing.

Credits and the audit log

What your agent's work costs comes out of the organisation's credits, as it would in Typestream, and the monthly cap applies. The settings page shows each client's calls, when it was last used and the credits it has used.

Every call is in the organisation's audit log, as the client acting for you, with whether it asked first. Anything it changes, such as a monitor, is logged as well.

Revoke access

On Settings → Connected agents, choose Revoke beside a client. It stops at its next call and can't renew its access; connecting it again means signing in again. You can revoke your own clients; owners and admins can revoke anyone's.

If it can't connect

  • "Connected agents are off for this organisation": an owner or admin needs to turn them on.
  • It asks you to sign in again: its access was revoked, or you've left the organisation. Connect it again.
  • It works in the wrong organisation: revoke it, connect again, and pick the right one when asked.