Build a scorecard by describing it

You can now connect an AI assistant to your DimensionKit account. Paste one URL into Claude, Cursor, or any other client that supports remote MCP servers. There are no API keys to create.

app.dimensionkit.com/mcp

The client sends you here to sign in, then asks what it may do: read your scorecards, or read and write them. A read-only connection cannot change anything. Every connection acts as you and only reaches your own scorecards.

If you do not have an account yet, start the connection anyway. You land on sign-in, register, verify your email, and come back to the approval screen.

Connecting an assistant is available on every plan, including the free open beta.

Describe the framework

The admin is still there. Dimensions, questions, scored answers, tier bands — click through, save, done.

If the framework is already in your head, you can skip that. Tell the assistant what you measure, what the questions should sound like, and what each band means. It creates a draft, adds the dimensions, writes the scored questions, sets the bands, and waits. A draft is not public. You read it, then you tell it to publish.

A few cases where talking is faster than the blank admin:

• A diagnostic you already teach. The questions live in a slide deck or a PDF. You describe the dimensions and the answer scale; the assistant builds the scorecard.
• Workshop or cohort intake. You need a short pre-work assessment before the first session. Describe what you want to know about each person and ship it the same afternoon.
• A maturity model with a lot of questions. Thirty scored items and a band for each dimension is a lot of form filling. Talking through it is less of that.

You still own the scoring. The same answers always produce the same result. The assistant builds the scorecard. It does not score the people taking it.

Ask what the submissions are saying

Give a connection read access and you can ask things you would otherwise export a CSV to answer.

• Which dimension is weakest across this month's leads?
• Who scored low on the thing we sell?
• What did people actually pick on question X?

The assistant can pull submissions, the answers, and the per-dimension scores. That is why a scored lead is more useful than a nameless PDF download: you can ask what they said.

Change a live scorecard without rewriting history

You can edit a published scorecard. Add a question, reword one, drop a dimension. Submissions already collected keep the scores and the wording they were given. History does not move under you.

When a scorecard already has submissions, the tools tell the assistant how many, so it can say what a change would affect before it makes one.

Cut it off

Connected apps in the admin lists every assistant attached to your account. Disconnect ends access immediately.

Removing the server inside Claude or Cursor only deletes that client's copy of the token. Disconnect from DimensionKit if a laptop is gone, or you just want it off.

Try it

Connect a client has the URL and the consent screen. What an assistant can do is the full list. Building by hand in the admin still works. None of this is required.