Skip to content

Ask your data anything.

Connect Claude, ChatGPT, Cursor, or any MCP client to Zappush and ask business questions in plain English: who your best customers are, who is about to lapse, who bought twice from a campaign. Answers run on live first-party data Zappush unifies server-side, so the number your assistant reports is the number you would activate on.

Book a Demo

From $59/month · 14-day free trial · Cancel anytime

Aisha Khan

Store owner, not analyst

MCP
  • Asked“Who's at churn risk?”
  • Answer23 customers
  • Time to answer1.2s
  • SQL written0
0 SQL
get answers to real business questions without writing a query or waiting on an analyst
Any MCP client
ask from Claude, ChatGPT, or Cursor with no custom integration to build or maintain
2 key scopes
read-only to explore safely, read-write to let the assistant build and sync audiences

Your answers are trapped behind a dashboard

The questions that move a business sit one join away from an answer: which customers are worth re-acquiring, who is about to lapse, which campaign produced repeat buyers. The data to compute them is scattered and incomplete. Browser pixels miss an estimated 30-40% of conversions to ad blockers, Safari ITP capping client-set cookies at 7 days, and consent gaps, so any "best customer" list built on pixel data already undercounts spend. Offline orders, subscription renewals, and CRM revenue usually never reach the analytics tool at all. So the only people who can answer a real question are the ones who can write SQL against the warehouse, and by the time a report comes back the decision window has closed. You end up re-acquiring customers who would have bought anyway, discounting Champions who never needed the incentive, and letting at-risk customers churn silently because no one queried for them in time.

How it works

  1. Zappush unifies your data server-side

    Web events, CRM records, lead forms, offline purchases, and subscription renewals all land in one first-party dataset through Zappush's server-side pipeline. The numbers behind any answer are complete, not the fraction a browser pixel survives.

  2. Connect your AI assistant over MCP

    Zappush exposes a Model Context Protocol server, the open standard for connecting AI assistants to live data. Point Claude, ChatGPT, Cursor, or any MCP client at it with a scoped key, and your assistant can see your customer, order, and campaign data. No SQL, no export, no bespoke integration.

  3. Ask questions, get decisions

    Ask in plain English: "Who are my top 50 customers by lifetime value?", "Which customers bought twice from the Black Friday flow?", "Who is at risk of churning this month?" The assistant queries live data and answers. With a read-write key it can go a step further and build or sync the matching audience to Meta, Google, or Klaviyo.

What you get

  • MCP-native, works with any assistant

    Zappush speaks the Model Context Protocol, so it plugs into Claude, ChatGPT, Cursor, or any MCP host without a custom integration. Your data becomes something you can talk to from the AI tools you already use.

  • Plain-English business questions

    Ask for best customers, churn-risk cohorts, campaign repeat-buyers, LTV rankings, or reorder timing in the words you would use with a colleague. No SQL, no dashboard filters, no waiting on the data team.

  • Answers from complete live data

    Every answer runs on the same unified first-party dataset Zappush captures server-side: web, CRM, offline, and subscription revenue. So "top customers" reflects real spend, not the 30-40% a pixel happens to see.

  • Scoped read and read-write keys

    Issue a read-only key for safe exploration, or a read-write key that lets the assistant act. It can build a churn-risk win-back audience and sync it to Meta or Klaviyo. Least-privilege scopes and audit logs apply on every call.

  • Decisions, not just dashboards

    The output is not another chart to interpret; it is the specific customers, the segment, and the next action. Ask a question, get an answer you can act on, and optionally have the assistant activate the audience in the same breath.

Use cases

FOUNDERS

Ask growth questions without SQL

The questions that move a business sit one join from an answer and behind whoever can write SQL. Connect Claude, ChatGPT, or Cursor over MCP with a scoped key and ask in plain English: top 50 customers by lifetime value, who is at risk of churning this month. The assistant queries live first-party data.

AGENCY REPORTING

Answer client questions from complete data

Pixel-built customer lists already undercount spend, and offline, CRM, and subscription revenue usually never reach the analytics tool at all. Every answer runs against the same unified server-side dataset covering web, CRM, offline, and renewals. The number your assistant reports is the number you would activate on.

GROWTH OPS

Turn an answer into a synced audience

Most analytics stops at a chart someone still has to act on. With a read-write key, the assistant builds the cohort it just described and syncs it to Meta, Google, or Klaviyo in the same exchange. Read-only keys stay available for safe exploration, with least-privilege scopes and audit logs on every call.

Questions, answered.

Do I need to know SQL or set anything up in a dashboard?
No. You ask questions in plain English from an AI assistant you already use, like Claude or ChatGPT. Zappush's MCP server handles turning your question into a query against your live data and hands back an answer. There's no query language to learn and no dashboard to configure.
Can the AI change or delete my data?
Only if you let it. Access is governed by scoped keys: a read-only key lets the assistant look at and analyze your data but never mutate it, so exploration is structurally safe even if a question is malformed or a prompt-injection attempt slips in. A read-write key additionally lets the assistant act, for example, build a churn-risk audience and sync it to Meta or Klaviyo, and every tool call is audit-logged.
How is this different from the AI chat inside my analytics tool?
Two ways. First, the data: most in-app chat answers run on whatever the pixel captured, which undercounts purchases by an estimated 30-40% and usually misses offline and CRM orders, so 'best customers' can be wrong. Zappush answers from complete first-party data it unifies server-side. Second, it's MCP-native and open, so you ask from the AI tools you already use instead of being locked inside one vendor's chat box, and with a read-write key the answer can become a live, syncing audience.

See what your data is missing.

Start free in minutes, or see it live on a call.

Book a Demo

From $59/month · 14-day free trial · Cancel anytime