# Apify Insights MCP

Case study by Devansh Tiwari (https://www.devanshtiwari.com) · https://www.devanshtiwari.com/work/apify-insights-mcp

Open-source MCP server that lets Apify creators ask Claude, Codex or Cursor about their own account: what earns, what broke and what to fix.

## Key facts

- Role: Creator, sole engineer
- Client: Open source
- Year: 2026
- Status: Shipped
- Source code: https://github.com/Devansh-365/apify-insights-mcp
- Tech: TypeScript, MCP SDK, Node.js, esbuild, Vitest

## At a glance

- Problem: Account answers were scattered across Apify's console, and AI assistants couldn't see them.
- What I built: A read-only MCP server that plugs into Claude, Codex, Cursor and other AI apps.
- Result: Ask "what broke last night?" and get the answer from live data.

## Numbers

| Metric | Value |
| --- | --- |
| Tools | 12 |
| AI apps supported | 6+ |
| Tests | 200+ |
| Releases in 3 days | 7 ([source](https://github.com/Devansh-365/apify-insights-mcp/releases)) |

Values with a + are public figures rounded down.

## The problem

The answers I needed every morning were spread across Apify's console: what earned, which runs failed and why, where users dropped off, how I ranked in the store. Getting them meant a lot of clicking, and an AI assistant couldn't help because most of that data has no public API.

## What I built

An MCP server that connects an AI assistant to an Apify account, read-only. It has 12 tools covering earnings, break-even prices, failed runs with log lines, user drop-off, quality fixes and store rank. One setup command adds it to Claude Code, Codex, Gemini CLI, Cursor and other AI apps, and a small browser bridge passes a short-lived login token so nobody copies tokens by hand.

### Morning brief

Ask "how am I doing?" and get earnings, failed runs and the scrapers that need attention in one answer.

### Why runs failed

Failed runs per day, the last lines of each log, and the build that was live when the failures started.

### Pricing help

Profit per scraper and the break-even price, so you can see which ones lose money and what price fixes that.

### Growth and rank

Where users drop off, which scrapers are rising or fading, and where you rank in the store.

### No token copying

A small browser bridge hands over a short-lived login token only when the server asks for one, saved in a file only you can read.

### Built for AI assistants

Every tool is marked read-only, a glossary explains each metric, and missing data shows as "not available" instead of zero, so the assistant doesn't guess.

## Highlights

- Ask in plain words, like "what broke last night?", and get answers from live account data
- One command adds it to Claude Code, Codex, Gemini CLI, Cursor and more
- It can only read. A fixed list of allowed requests means it can't change or delete anything
- 200+ automated tests and 7 releases in 3 days
- The same pattern works for a client: give your team's AI assistant safe, read-only access to data it can't reach today

## Work with Devansh

Devansh Tiwari is an AI product engineer based in India, available for freelance and consulting work: custom data feeds from sites without APIs, AI features (agents, chat, RAG, MCP servers) built on that data, and running them after launch.

- Book a 15 minute call: https://cal.com/devansh0718/15min
- Email: devanshtiwari365@gmail.com
- All case studies: https://www.devanshtiwari.com/work
- Site summary for AI tools: https://www.devanshtiwari.com/llms.txt
