How to Connect Promptwatch MCP to Claude, Cursor, and ChatGPT

Analysts already work in Claude, Cursor, or ChatGPT. We do not make them live in the Promptwatch UI for every check. We add the hosted MCP server and they ask the live project. Same data as the login. Fewer screenshots in Slack.

Essential at $95/mo includes MCP and API. That is the first brand plan where we bother. Explore is free, 10 ChatGPT prompts, and is not an MCP rollout. You can learn the URL there. We do not staff a retainer on it. Professional is $245/mo. Business is $579/mo. Kick-off is $199/mo, unlimited projects and prompts, 10 seats. G2: 4.7/5. Confirm Essential, Professional, Business, or Kick-off so MCP and API are on the plan. Pay for Essential or Kick-off before we add MCP to everyone's editor.

What we add

MCP is hosted HTTP. There is a server URL. We paste that URL into Claude, Cursor, or ChatGPT. Auth is OAuth or an API key (Bearer). We do not run a local MCP process on analyst machines. If their laptop dies, the server is still there.

The ChatGPT plugin and the Claude connector use the same server. Cursor gets the same URL. We are not maintaining three products. If a setup works in Claude and fails in Cursor, the usual miss is the key scope, not a different endpoint.

OAuth is fine when the person has a Promptwatch login. API key is what we use on shared analyst setups and in Cursor so we can rotate without chasing Google grants. Read-only either way.

Keys come in project, org, and read-only flavors. We use read-only keys on analyst laptops. Project key when they only touch one logo. Org key on Kick-off when they jump accounts. Write keys stay off these machines. The CMS-connected seat is the only place we keep a write key, and only because publish needs it. See read-only keys.

Create the read-only key first, project or org, matching who will chat. Then add the hosted MCP URL in Claude, Cursor, and ChatGPT. Same URL. OAuth or the key. Ask for mention vs citation on a named prompt and confirm the answer matches the UI before you call it done. That one check catches a bad scope faster than any setup guide.

What we ask it

After GSC import, we ask which imported prompts are mentioned vs cited. That workflow is GSC topics with MCP.

Without GSC, we still ask for the last week of visibility, the URL that got cited, and whether a persona row looks different from the default. We do not ask it to invent a client result. We do not ask it to average engines. Google's AI optimization guidance still applies to the pages those citations point at. MCP reads the scoreboard. It does not write the page.

MCP does not replace crawler logs or the review inbox. If the question is "did ChatGPTBot fetch this URL," we open Agent Analytics. If the question is "can we ship this draft," a human accepts in the inbox. Chat is for reading the scoreboard faster.

What we will not wire

We will not put a write key in a shared Cursor config checked into git. We will not add MCP to a contractor laptop on a one-week audit without a read-only project key we can revoke.

We will not tell a client that ChatGPT's plugin is a different dataset than Claude. Same server. If the answers disagree, the prompt or the model on top changed, not the Promptwatch project.

No instant alerts through MCP. Scheduled Slack briefings are a different integration.

Revoke laptop keys when someone leaves the account.

FAQ

Is the ChatGPT plugin a different dataset than Claude?

No. Same hosted MCP server. If the answers disagree, the prompt or the model on top changed, not the Promptwatch project.

Which plan includes MCP?

Essential at $95/mo is the first brand plan where we bother. Explore is free, 10 ChatGPT prompts, and is not an MCP rollout.

Can we put a write key in a shared Cursor config?

No. We will not put a write key in a shared Cursor config checked into git. Read-only keys on analyst laptops.

If you want the URL and key issued, hello@1001seomedia.com. Say which editors you use. Do not send a write key in email.