Connect an AI Client (MCP)
Connect Claude, ChatGPT or another MCP-capable client to your Noise data: ask questions in plain language and manage campaigns with confirmation.
Noise exposes your brand's data over MCP (Model Context Protocol), so an AI assistant like Claude or ChatGPT can answer questions about your campaigns directly, "how did last week compare to the week before?", and even manage campaigns and playbooks for you, with a preview and confirmation before anything changes.
The server URL to add to your client:
https://reporting.getnoise.com/mcp
Connecting Claude
- In Claude, open Settings → Connectors, add a custom connector, and paste the MCP URL above. Other MCP-capable clients work the same way: add a remote MCP server by URL.
- When prompted to sign in, use the email you use for the Noise portal. Your assistant is linked to your brands automatically, and you are done.
- Signed in with a different email? Ask the assistant for your reporting data anyway. It will reply with a connection code starting with
org_. Paste that code on the Integrations page (Settings → Integrations, or Resources → Reporting API on the classic dashboard) and click Connect.
Connecting ChatGPT
- In ChatGPT, open Plugins and search for Noise for Brands. If it isn't listed for your account, choose Add → Add custom MCP server, name it Noise for Brands, paste the MCP URL above, and keep OAuth as the authentication. Custom MCP servers need a ChatGPT plan that allows them.
- Click Connect and sign in with the email you use for the Noise portal. As with Claude, your assistant is linked to your brands automatically; if you signed in with a different email, use the connection code described above.
- In a new chat, turn on Noise for Brands from the + menu and ask away.
- ChatGPT asks your permission before each change. Allow once approves just that change; Always allow stops asking for every later change from Noise for Brands.
New to Noise?
You can try Noise from your assistant before you have an account. Add the server URL above to your assistant, sign in with the email you want on your Noise account, and ask: "Help me get started on Noise."
Your assistant can explain how Noise works and why many creators beat a few, explain billing (you pay per 1,000 views, nothing upfront), define any Noise term, show you real creator posts in your niche, and help you plan your first playbook.
When you are ready, create your brand in the Noise brand portal with the same email. Your assistant links to the new brand automatically. You can finish the remaining setup in the portal, in your assistant, or a mix of both: the assistant can tell you what is left with one question, "what does my brand still need?"
Started in the portal instead? The onboarding screens offer Onboard with your agent at the top, and the step right after your brand is created asks whether you want to continue here or with your assistant.
Managing linked clients
The Integrations page lists every AI client connected to your brand and which teammate connected it. Anything to know:
- Every teammate can connect their own client. Links are additive and per-person.
- One client, several brands. If you work across multiple Noise brands, the same AI client can be linked to each of them; the assistant will ask which brand to query.
- Remove anytime. Removing a client on the Integrations page revokes its access within about a minute.
- New tools appear on reconnect. If we ship new capabilities, remove and re-add the connector in your AI client to pick them up.
What the assistant can see and do
Reading: daily performance (spend, views, posts, creators, CPM, installs), campaign and playbook details, the Meta and TikTok ads you run from Noise posts, and the change audit log. The same numbers, computed the same way, as your dashboard and the Reporting API; see Understanding Your Metrics.
Managing: the assistant can also pause or activate campaigns, adjust budgets and daily targets, update rates, and create, edit and publish playbooks, including full slide content, captions, hashtags, reference videos, and slide images from your image library. Three guardrails apply to every change:
- Preview first: the assistant shows you what will change, and your AI client asks your permission before anything is applied.
- Sensible limits: changes are bounded (for example budget and CPM caps), and write operations are rate limited.
- Attribution: every change is attributed to the teammate whose connection made it, and shows up in your audit log like any dashboard change.
Billing, payment methods, Noise Unlimited, and launching paid ads stay in the Noise portal; the assistant links you to the right page. It never takes card details in chat.
The assistant only ever sees the brands it is explicitly linked to, and access is scoped to your organization's own data.
Run the flywheel on a schedule
Noise works best as a loop. You brief creators with a playbook, creators find your campaign, you find the posts that win, and you amplify those winners by refining the playbook or running them as Meta or TikTok ads from the Noise portal. Then the loop starts again with what the winners taught you.
Ask your assistant "how do I get the most out of Noise?" and it walks you through the loop with your own numbers. It then offers to run the loop for you on a schedule, for example four times a day. Say yes and it sets up a recurring task in your AI client that reads your recent posts, picks the winners, suggests playbook changes, and reports back.
Each scheduled run previews its changes and waits for you. Playbook edits are only applied when you confirm them in your next conversation. If you would rather let it act on its own, tell the assistant so and it adds that permission to the task, within the usual limits.
Prefer raw API access?
If you want to pull numbers into your own tools rather than chat with them, use the Reporting API. The MCP server and the API expose the same reporting data.
Need help?
Email [email protected]. If you already work with a Noise rep, your assistant can give you their email too: ask "who is my Noise contact?"
Sign up with Google