Connected agents
Claude, ChatGPT, Cursor or your own agent can work in Pluto Ads as you, and never with more access than you have.
Connect by signing in
This works for Claude, ChatGPT, Cursor, Claude Code, VS Code, Codex and other MCP clients that sign in. There's no key to copy.
Access starts at Full access: everything your role allows in that workspace, never the organization. The line under the choices says what the one you picked leaves out, like "Can't publish or change live ads." Choose View results or Build and publish to give less, or Custom to tick lines yourself: the choices are in Access. Access lasts 90 days, or less when your organization sets a shorter limit, and works in every ad account of the workspace. Under Custom you also choose its Ad accounts (all of them, or the ones you tick) and when it Expires. An agent with chosen ad accounts sees and changes only what's in them; see Ad accounts.
With Full access and an admin role, an agent can add and remove people and change settings, as you could. Choose less for an agent you don't trust with that.
When your organization lets members' agents only view, the screen says so and offers View results only. When it lets them build and publish at most, Full access isn't offered and the line under Build and publish says so. When it limits how long credentials last, Never isn't offered. See Security.
Cancel shares nothing and shows Not connected. A sign-in you don't finish expires within an hour: start again from your client.
Who can connect agents
- Organization owners and admins can always connect agents for themselves. With one workspace, these are its owners and admins.
- Members, and admins of only one workspace in an organization with several, can while Members can connect agents is on. It's on by default, in SettingsGeneral (Settings, then General), or in the organization's General settings when it has several workspaces. Turning it off stops their agents and agent credentials on their next request.
- Guests can't connect agents.
- In the organization's Security settings, owners and admins can let members' agents only view, or build and publish at most. That applies on each agent's next request. Owners and admins aren't bound by it.
An agent follows your role on every request. If your role changes, the agent's access changes with it. If you're suspended or removed, every agent you connected stops.
What agents can do
An agent does what you do in the app, through the same actions and checks. With the access you gave it, it can:
- Read and search launches, ads, media, results and the Meta and LinkedIn and X connections.
- Create and edit launches, ad sets and ads in bulk. An ad set it fills past Meta's usual 50 ads keeps its ads, and the agent offers to split it if you want.
- Import and organize media, and search it. With Creative Memory on Enterprise, search inside your creatives.
- Comment, review, change statuses, tags and assignees, and notify people.
- Build, test and run loops.
- Propose several live changes as one plan you approve in the Inbox, like the Agent does.
- With publish access, publish launches and change what runs in Meta or LinkedIn or X: delivery, budgets, bids, schedules and targeting. In LinkedIn that covers what's in a launch: import a campaign made in Campaign Manager first.
- With admin access, manage members, API keys, webhooks and settings.
- On X, work with IDs, your launches and what Pluto Ads did. X's terms keep what X reports (names, statuses,
results, leads' answers, replies) in the Pluto Ads app until X agrees in writing that it may leave, so an agent
gets those fields as
nullwithwithheld: true. See What stays in Pluto Ads.
Every tool is in the tool catalogue. Some steps always stay with a person: approving at Facebook or LinkedIn or X when you connect them, paying, and changing ownership.
Publishing with an agent
Publishing creates campaigns, ad sets and ads in Meta or LinkedIn or X that spend money. An agent needs publish access, and the publish is confirmed in your conversation with the agent:
Pluto Ads can't see your conversation, so it can't tell whether you said yes. Give publish access only to agents you trust to ask. For everything else, leave the Live ads lines unticked. See Publishing.
What your team sees
- Presence. An agent in a launch shows in the launch header with the Agent mark, next to the people there.
- Names. A connected agent is named after the app and you, like "Claude via Sarah Mitchell". Activity, comments and the Inbox use that name, so your team knows it wasn't you in person.
- Audit log. Admins see every change in SettingsAudit log (Settings, then Audit log), with the agent and the person it acted for.
Anything an agent reads in your workspace, like ad copy, file names and comments, is content, not an instruction to the agent.
See, narrow and disconnect agents
SettingsAPI keys and agents (Settings, then API keys and agents), under your account, lists your agents in every workspace, beside your API keys and agent credentials, with what each can do, its ad accounts ("All ad accounts", or their names), its Workspace, when it was Last used and from which address, when it was Created and when it Expires.
- Narrow access... gives an agent less access without connecting it again. It opens the same choices as when you connected it, offering only what the agent can do now and the ad accounts it works in now. Select Save: from its next request it does only what you kept. Narrowing never adds access; to give an agent more, connect it again.
- Disconnect agent..., then Disconnect, ends its access right away. The API keys and agent credentials it made are revoked with it, and the webhooks it made are turned off.
Admins see every agent in the workspace in SettingsAPI (Settings, then API), under Connected agents, with who each Acts for. They narrow any of them the same way, and Revoke access stops one, with what it made.
Good to know
- An agent works in the one workspace you connected it to. To work in another, connect it again and choose that workspace.
- Agents share the workspace's rate limits with API keys. See Rate limits.
- To run your own agent or service without anyone signing in, an admin creates agent credentials.