AI connectors
Your clients’ findings, in Claude and ChatGPT
Yellowhead Digital is the intelligence layer for your marketing stack. Connect Claude or ChatGPT to it by signing in, and your assistant can read your Diagnostics, reports, scans, action items and Portfolio, and start a Diagnostic when you ask.
It works on every plan, Free included. What it can read follows your plan.
What you can ask
Written for agencies working across a book of clients. Consultants and fractional marketing leads do the same work solo on the Consultant plan, one of the Agency plans. Running your own brand? The report, scan, allowance and Diagnostic questions work on your own account too.
Which clients need attention
“Which client workspaces moved since last month?”
The assistant reads the Portfolio: each client’s latest MIDAS (Marketing Infrastructure Diagnostic Audit Score), how it changed, and the open action items marked critical or high. Moves smaller than the usual run-to-run variation are left out, so a client shows up because something changed.
Team accounts on an Agency plan, Business Enterprise or a custom plan
Before a client call
“Summarize the latest report for this client and its lowest-scoring pillars.”
It reads the report as your account sees it: the ten pillar scores as the business rated them and as Yellowhead detected them, the executive summary, and the blind spots. A Free plan report carries less detail, the same as in the app.
Every plan
What to fix first
“Which action items are open on the latest report?”
Each action item comes with why it matters, what to do, its priority, its status and the Diagnostic or scan it came from.
Pro, Enterprise and every Agency plan
What a client’s scan found
“What did the latest Crawlability Scan find on this client’s site?”
It reads the account’s Citation, Crawlability and Accessibility Scans: which websites AI answers cite, what robots.txt lets each AI crawler reach and how pages answer AI crawlers compared with a browser, and failing automated accessibility checks by page template. Page-level detail stays in the app.
Every plan, for scans the account has run
A prospect before the first meeting
“Start a Diagnostic of example.com and file it as a prospect.”
The assistant shows a preview first and nothing starts until it is confirmed. On an Agency plan, an owner or admin of the team account can have it create the prospect or client workspace in the same step.
Every plan starts Diagnostics; new workspaces need an Agency or custom plan
This month’s allowance
“How many Diagnostics do we have left this month?”
How many included Diagnostics are used, how many remain, and the date the allowance resets.
Every plan
What a connection can do
The consent page lists what a connection will be able to do in the account you pick, before you approve. The list follows that account’s Yellowhead plan.
| What it can do | Which accounts |
|---|---|
| Read the Diagnostics and reports in the account | Every plan |
| See the account’s plan usage | Every plan |
| Read published Strategic Assessments and Infrastructure Audits | Every plan |
| Read the account’s Citation, Crawlability and Accessibility Scans | Every plan |
| Read the account’s action items | Pro, Enterprise, every Agency plan (Consultant, Studio or Firm) and custom plans |
| See the Portfolio: your client workspaces and how their scores moved | Team accounts on an Agency plan, Business Enterprise or a custom plan |
| Start a Diagnostic using one of the account’s included monthly Diagnostics | Every plan. On a plan with workspaces it can file the run in one; on an Agency or custom plan it can also create a new prospect or client workspace when you ask |
A connection made before Yellowhead added scan reads on 2026-11-06 reads scans only after you reconnect it. Creating a prospect or client workspace needs you to be an owner or admin of the team account. The tools reference lists every tool the agent door serves.
Starting a Diagnostic from a chat
Asking for a Diagnostic starts with a preview, and the preview starts nothing. It shows the website, the business name, the self-ratings, where the run will be filed, and which of your included monthly Diagnostics it uses. The run starts only when the assistant confirms that preview, and any change needs a new preview.
An assistant may confirm without asking you again. If you want to approve each run, ask it to show you the preview and wait for your go-ahead.
A run uses only your account’s included monthly Diagnostics. When none are left, nothing is charged, and the assistant is told the reset date. One person can start only so many from an assistant in a day. A Diagnostic takes 5 to 10 minutes, and an email arrives when it completes.
What stays in your control
What a connection can and cannot do, and how to end it.
You approve it on our consent page
You sign in on app.yellowhead.digital. The consent page names the app when we can verify it, labels one that registered itself automatically as unverified, and shows where you return to afterwards. You pick the account it can read and see the list of what it will be able to do there before you choose Allow or Deny.
It reads, and starts Diagnostics when asked
A connection made by signing in cannot change, comment on or close an action item, on any plan. The one thing it starts is a Diagnostic, a new one or a re-check of a site from an earlier report, after a preview and only from the account’s included monthly Diagnostics. When none are left, nothing is charged.
Your permissions, in one account
A connection can do only what you can do in the account you connected it to, and it reaches that account only. To use another account, connect again and pick it.
Remove it whenever you like
Settings › AI assistants in the app lists each assistant you have connected, the account it reads and what it can do. Remove cuts off its access within about a minute. A connection that goes unused for 30 days ends on its own.
Team owners and admins decide
Owners and admins of a team account see everyone’s connections under Access › AI assistants and can remove any of them. Turning off “Members may connect AI assistants” ends members’ connections and stops new ones. Someone with client access to an account cannot connect an assistant to that account yet.
What it reads goes to the assistant
Whatever the connection reads is sent to the assistant and kept under that app’s own policies, as the consent page says.
When an agent should request changes
A chat connection reads and starts Diagnostics. For an agent that works your action items over the Model Context Protocol (MCP), Enterprise and every Agency plan include MCP Agent oversight. A workspace owner or admin sets up the agent’s access one capability at a time, which can include updating and commenting on action items. Every change the agent requests is held for a person to approve, unless the workspace opts that access into applying changes directly.
Set up the connector
The connector is Yellowhead Digital’s MCP server, the agent door, at https://api.yellowhead.digital/mcp. You add it to
your assistant yourself and sign in with your Yellowhead Digital account.
Claude
Add Yellowhead as a custom connector in Claude’s connector settings, switch it on in a chat and ask “Using Yellowhead, …”. Claude asks you to sign in the first time a question needs your account.
Connect Claude →ChatGPT
Add Yellowhead as a custom MCP server with OAuth in ChatGPT, sign in, then mention @Yellowhead in a new conversation.
Connect ChatGPT →- You need a Yellowhead Digital account. A Free account works.
- Your Claude or ChatGPT plan has to allow custom connectors. Anthropic and OpenAI decide which plans do.
- Building your own client? The agent-door documentation covers authentication, capabilities and rate limits.
Start with the free Diagnostic
A connection reads what is in your account. Run the free Diagnostic on a client or prospect, then connect your assistant and ask about it.