Guide · MCP connector
How do you use your AI visibility results from your own AI assistant?
Short answer
Connect the CDE AI Visibility Advisor to your AI assistant once, through its MCP connector. Then ask in your own words: what to focus on this week, who AI answers name instead of you, what customers complain about. The Advisor measures in the background. Your assistant reads the published results and answers with the release date and the evidence. A routine can ask the same questions every Monday and report the summary.
Your AI assistant is the front end. The Advisor works in the background. Results, plan, evidence and briefs come through your assistant; team invitations and parts of the settings stay in the portal. Included with Start, System and Scale.
Updated 16 November 2026 · By Balázs Turán
See it in 49 seconds
Real read-only results for Creative Data Engineers. Steps 1 and 2 re-enact the dialogue with real results from 14 November 2026. In step 3 the setup is re-enacted; the summary is the real routine run of Monday 16 November 2026, 08:09, with nobody at the keyboard. The video has no sound. The transcript has the same text.
Read the transcript
Opening screen: Ask your own AI assistant. It answers from your AI visibility results.
Step 1, ask in your own words. You: "What should we focus on this week?"
Step 2, your assistant looks it up. Claude: "Your latest results are from 13 Nov 2026. This week: make Creative Data Engineers identifiable to the AI (to do); page on the best AI visibility agencies in Germany (done); page on the best GEO agency (done). Next week: buyer questions about AI search visibility services and how an AI visibility score is calculated."
Follow-up. You: "Who do the AI answers name instead of us for the score question?" Claude: "For 'How is AI visibility score calculated?' you appear in none of the 8 measured checks. Named instead: Peec AI (3 answers), Semrush (3), AIghty20 (2) and others. Channel to publish on, from the cited sources: LinkedIn. Suggested: a page that gives the answer first, then the formula and one worked example."
Step 3, let a routine ask for you. You: "Ask this every Monday at 8:00 and send me the summary." Claude: "Done. Routine: every Monday at 8:00, read-only. Anything that changes your account, like marking a task as done, waits for your yes."
The routine runs on its own, Monday 16 Nov 2026, 08:09. Nobody typed anything; the summary arrived as a notification and a file: "No new release since 13 Nov 2026. Results change once a month. Top open actions: How is AI visibility score calculated? What is AI search visibility? What is the difference between GEO and SEO? In each, you are absent from 8 of 8 measured checks. Writes performed: none."
Closing screen: Connect once. Ask by hand. Then let a routine ask for you. Anything that changes your account waits for your yes.
Three steps
- Connect once.
In the portal, open Account settings, then MCP connector. Copy the server address into your assistant's custom connector setup and sign in with your portal account. Access follows your role in the account.
- Ask by hand.
Start with five questions:
- What should we focus on this week?
- Show the evidence behind recommendation 1.
- Draft the page from this week's brief.
- Check this page before we publish.
- Mark recommendation 1 as in progress.
The last two change something in your account. Your assistant shows what will change and waits for your yes.
- Let a routine ask for you.
Once the answers are useful, schedule the same question. Our own routine is a scheduled task in the Claude desktop app, every Monday at 8:00. It reads the weekly plan, reports "no new release" when nothing changed and delivers the summary as a file and an app notification. It runs while the app is open. A routine reads and reports. Anything that writes waits for a person.
What you can ask
Weekly focus
Ask "What should we focus on this week?" and your assistant reads the weekly content plan, the top open actions with the reason for each and what "done" means for each. The video above shows a real answer and a real Monday routine run.
Competitor changes
Ask "Who do AI answers name instead of us, and what changed since last month?" Your assistant reads which brands the measured AI answers name for each of your questions and which sources they cite. The month-over-month comparison needs two monthly results that can be compared. Our own account does not have that yet, so we show no example report here. We will add one when it exists.
Voice of Customer
Ask "What are customers negative about right now, and what should we answer first?" Your assistant reads the weekly Voice of Customer results: topics, sentiment, quotes and changes from public discussions such as YouTube and Reddit. Collection runs weekly once you set up search terms and sources.
What your assistant does and does not do
Does
- Reads your published results, evidence, briefs and schema drafts.
- Plans the week from the content plan.
- Checks a page before you publish it, after asking you.
- Marks progress on a recommendation, after asking you.
- Names the release and the date it reads from.
Does not
- Start a new measurement. AI answers are collected automatically on a fixed schedule and evaluated every month; Voice of Customer runs weekly.
- Invite teammates. Team invitations and parts of the settings stay in the portal.
- Read results that are not published yet.
- Promise rankings, citations or recommendations by AI services.
Which AI assistants work
Claude, ChatGPT, Microsoft Copilot and Codex. The connector uses the open Model Context Protocol (MCP). Every assistant that supports remote MCP servers can connect: add the server address from Account settings as a custom connector and sign in with your portal account.
What stays in the portal
The portal shows the same results as a dashboard, if you prefer to look. Team invitations and parts of the account settings stay in the portal.
Questions about your AI assistant
Do I still need to log in to the portal?
For team invitations and parts of the settings, yes. For your results, plan, evidence and briefs, your AI assistant is enough.
Can my assistant change anything without asking?
No. Before any change it shows what will change and waits for your yes. A routine only reads and reports.
How often is there something new?
Every Monday: Voice of Customer, Search Console and the open actions for the week. AI answers are collected in several rounds during the month. Once a month they are evaluated together with the readiness check and published as a new report with new prioritized recommendations. Between monthly releases your assistant tells you that no new release was published and which actions are open this week.
Which plans include the MCP connector?
Start, System and Scale.
Can it start a measurement on demand?
No. AI answers for the questions you agreed are collected automatically on a fixed schedule and evaluated every month.
See where your brand stands
Check one page for free with the AI Visibility Inspector, or start with a plan.