Ask Claude. It answers from your broch.
Every channel cleaned, joined and defined once, plus what you told us about the business. Claude reads all of it before it answers, in the tool your team already has open.
In beta today, on your own login, with Claude Code.
What your AI can ask for.
Seven tools, reading the same numbers your dashboards run on. The answer in Claude is the number on the wall.
Daily summary
“How did Google Ads do last month against the month before?”
Purchase funnel
“Where does the funnel leak most, view to cart or checkout to purchase?”
Custom query
“List every campaign whose spend rose but conversions fell in July.”
Search the Brain
“What does the strategy document say the priorities are this quarter?”
Document contents
“What sections does the launch brief have?”
Read a section
“Read the returns section of the delivery policy in full.”
Propose a data issue
“File the 12 June bot wave as a data issue for the agency to check.”
Prompts, ready to run
Type the slash and pick one.
The funnel and Brain tools appear once your broch has a funnel and a Brain to read.
It reads your business before it reads your numbers.
An AI on the raw APIs looks your data up. An AI on your broch is briefed first.
What you told us, first.
Your business profile, written by you on your settings page: what you sell, when you are busy, what you are working on. The AI reads it before any number. What you have not told us stays unknown, never invented.
A known problem is never news.
A bot wave doubled June’s sessions. Your team knows, and so does your broch, because it is written down beside the data. Every answer that touches that window carries the note.
It says which revenue it means.
Booked or paid: the dashboard reads one ladder and the AI quotes the same one, naming it whenever the two differ. If the data stops two days ago, it says so.
Read-only, and every call logged.
The rules, whatever it is asked.
Bring the tool you already use.
Claude Code connects today. Claude Cowork, ChatGPT and Codex follow, then anything that speaks MCP over HTTP.
Connected in three steps.
- 1 Install Claude Code.
- 2 In your broch, open MCP and copy a token. It is shown once.
- 3 Paste the command into a terminal, then ask.
claude mcp add --transport http broch https://app.broch.ai/mcp \ --header "Authorization: Bearer <your token>"
Prompts: type /mcp__broch__ in Claude Code and pick one. Replacing a token: run claude mcp remove broch first.
Fair questions.
What works today, and what it can’t do.
Can we use it today?
Yes, in beta, from your own login. On Lite you set it up yourself in the app; on Premium we set it up with you.
What does beta mean here?
Open to every client on their own login, and still improving. Check an answer against the dashboards before you act on a big one.
Which AI tools work?
Claude Code today. Claude Cowork, ChatGPT and Codex next, then anything that speaks MCP over HTTP. Nothing to install on our side: your broch is the server.
Can it change anything?
No. The one exception files a data issue for a person to verify, and until they do it changes nothing.
What gets logged?
Who asked, which tool, what they asked, how long it took, how much came back. Never the answer. Each token’s activity sits beside the button that revokes it.
Is it on every plan?
Yes. Self-serve on Lite; on Premium we set it up with you.
Why not plug Claude straight into Google Ads?
You can, one platform at a time, and each one marks its own homework. Your broch gives the AI the joined history, the definitions and the notes underneath. The five reasons, on the homepage.
Point your AI at the clean layer.
Email us and we’ll show you a broch answering in Claude, then talk about yours.
In beta today, on your own login.