Model Context Protocol

Ask your own AI for the verdict.

Connect Reckon to Claude, ChatGPT, Cursor or whatever you already work in. It reads your Meta and Google performance, what each ad actually says, and Reckon’s call on every one of them, judged against the other ads selling the same product.

https://usereckon.ai/connect/mcp

Four steps. No API key.

No Reckon account yet? Connecting makes one. Your assistant sends you to Reckon, you sign up and name a workspace, and you land back where you left off. The one thing to have ready is an ad account connected to that workspace, since that is what your assistant will be reading.

01

Add the Reckon URL

Paste it into your assistant's connector settings, or run one command.

02

Sign in, or sign up

Your assistant opens Reckon. Sign in, or create an account there and then; either way you come back to where you left off. No key to copy, no secret to store.

03

Pick the workspace

Choose which workspace to share. That choice is what the connection can read.

04

Connect an ad account, once

Connecting an assistant cannot connect Meta or Google Ads: an admin of the workspace does that inside Reckon, at https://usereckon.ai/connect. A new workspace has nothing to read until then, and your assistant will say so and point you there. The connection you just made keeps working; ask again once the first sync lands.

05

Ask your first question

Connecting is not the same as having data. Check which accounts it can see, how current they are, and how much creative Reckon has read.

  1. 1Open Customize, then Connectors, press the plus and choose Add custom connector.
  2. 2Paste the Reckon URL and add it.
  3. 3Press Connect, sign in with your Reckon account and pick the workspace to share.
https://usereckon.ai/connect/mcp

Works on every Claude plan, including Free with one connector. On Team and Enterprise an owner adds it first under Organization settings, then Connectors, and each member connects from Customize.

Claude's guide to custom connectors

Then ask this first

Which ad accounts can you read through Reckon, how current is each one, and how much of our creative has Reckon analysed so far?

A good answer names your accounts, the newest day Reckon holds for each, and how much of your creative it has read. If it does, you are ready for the questions above.

If it does not work

There is no workspace to pick
Your Reckon account is not in a workspace yet. Reckon asks you to name one on the way through; if you skipped it, or your team already has one, ask an admin to invite you.
It connects but lists no accounts
Nobody has connected an ad account to this workspace. An admin does that once, in Reckon.
The numbers look thin or old
The first sync may still be running. Ask your assistant for the sync status, or check again after tonight's refresh.
Sign in is refused
On a managed Claude, ChatGPT or Microsoft 365 workspace an admin may have to allow the connector first. See the note under your assistant's tab.

Access

It reads. It cannot touch your ad account.

You sign in to Reckon and choose one workspace to share, the way you would with any connected app. There is no API key to keep. The connection then reads that workspace through Reckon’s own tools, on the same limits as the dashboard.

It cannot touch your ad account

No tool can pause a campaign, move a budget, publish an ad or change a setting in your workspace. Three tools write, and only to Reckon: a hypothesis you ask it to track, a report that Reckon got something wrong, and an Ad Library link you ask it to fetch. Each is marked as a write, so your assistant asks you before running it. The server runs no AI of its own: it answers from what Reckon has already stored, and never asks Meta or Google for anything on your behalf.

The grant is one whole workspace

You choose the workspace when you sign in, and the connection can read all of it through these tools: performance, creative text, brand notes and boards. Naming an account in a question narrows that question, not the grant.

Removing a member ends their access

Reckon checks membership on every request, and clears what it remembered the moment someone is removed from the workspace, so they lose access on their next question. That is the fast way to cut someone off. Disconnecting the assistant stops new sign ins, but a token it already holds can keep working for up to a day.

Calls are recorded and bounded

Reckon records which assistant read what, and when. There is no screen for that yet, so ask and we will send it. Limits on how much one workspace can ask at once protect the dashboard, which shares its servers with these reads.

Your AI provider is still your AI provider

Answers travel to whichever assistant you connected, and what it does with them is governed by that product's own terms, not by Reckon's. Ending access stops future reads. It does not recall what was already delivered.

Questions worth asking

Each one is a decision a strategist has to make anyway. Your assistant can recommend a change. It cannot make one: nothing here can pause a campaign, move a budget or publish an ad, so what to act on stays your call.

  • Which ads should I turn off this week, and where does that money go instead?

  • Which product are we spending real money on where every ad selling it is weak?

  • Which ad is strong enough to build a whole batch around, and what should the next five versions keep and change?

  • Our best ad stopped working. What in the data is consistent with people tiring of it, what else could explain the drop, and what can the data not tell us?

  • Which ads lose people in the first three seconds even though the ones who stay end up buying?

  • Rank the angles Reckon has tagged in our last 60 days by cost per sale, say which rows are too thin to call, and name an angle that does not appear in that window at all.

  • Which messaging angles do the brands we track run that do not show up in our last 60 days? Give them to me as test candidates, and flag any factual claim a human has to substantiate first.

  • Write Monday's brief: what changed last week, what we should shoot this week, and the numbers that argue for it.