AI
Your AI assistant talking to your data, not to a sample of it
CrazyLeads exposes an MCP server. Connect Claude, ChatGPT or Cursor and ask in plain language — and it executes, under the same rules and validations as the interface.
One query, not three hundred calls
The assistant writes SQL and reads the result. Cheaper, faster and more accurate than crawling an API item by item.
It writes, not just reads
Creates artifacts, sets up experiments, configures deliveries — through the same validations as the UI.
Scoped and revocable
The token is yours, scoped to the active project, and can be revoked at any moment.
Asking
The question you would have filed with the data team
Instead of opening a ticket and waiting, you ask. The assistant discovers your project tables, writes the query, runs it and returns the number — and shows the SQL, so you can check where the answer came from.
You
How much did each Meta campaign sell in the last 30 days,
and what is the return on each one?
Claude · via CrazyLeads MCP
I joined Meta spend with approved checkout sales.
Return by campaign:
CAMPAIGN SPEND REVENUE ROAS
Launch – cold 18,240 71,980 3.9x
Retargeting – 7d 4,110 29,350 7.1x
Top of funnel 12,870 14,220 1.1x
Top of funnel is the only one below 2x.
Want me to publish this as a dashboard?Doing
From "show me" to "then go do it"
The useful part comes after the answer. The assistant publishes the dashboard, creates the experiment, configures the conversion delivery. And because the tools follow product rules, a delivery created by AI is born validated: the filter is tested against recent real events before anything goes live.
- Creating publishable artifacts and dashboards
- Configuring deliveries with validation and a dry run
- Creating and reading A/B experiments
- Querying leads, timelines and integrations
Governance
AI with permission, not with open access
Every token has a scope and a project. What the assistant can see is what you authorized, and every write passes the same validation as the interface. Revocation is immediate.
Frequently asked questions
Will my data be used to train models?
Not on our side: MCP is an authenticated read and write channel. What your chosen AI provider does with conversation content is governed by their terms — worth checking whether you are on a personal or business account.
Which assistants does it work with?
Any MCP-capable client. Today that includes Claude, ChatGPT and Cursor, among others.
What if the AI writes the wrong query?
You see the SQL. That is the difference between a number you can verify and a number you have to trust.
Start measuring what actually becomes revenue
Create a free account, connect your first source and see your first lead timeline today. No card required.