← All agents

The AI Watchdog

See exactly how your brand shows up when someone asks ChatGPT, Perplexity, Claude or Gemini about your industry, and which competitor is getting mentioned instead of you.

Comment WATCHDOG Needs: free Apify account ~5 min install MIT · 28★ Tested 2026-09-05

What it does

The AI Watchdog is an MCP server that checks how your brand actually shows up inside four AI search platforms: ChatGPT, Perplexity, Claude and Gemini. Point it at a brand name and it comes back with a mention rate, a position score, whether the AI is citing your domain, the sentiment of the mention, and which competitors are getting mentioned in the same answer. Run it across a few queries and it rolls all of that up into a single GEO score with plain English recommendations.

The claim in the video holds up. In testing, the repo cloned, installed and built cleanly, and starting the server confirmed exactly the two tools the README documents, audit_brand_visibility and check_brand_in_queries, nothing extra, nothing missing.

The honest caveat: this isn't a fully free tool once you actually run a check. It works by calling a paid Apify actor to query the four platforms for you, so you need your own free Apify account and API token. That's disclosed upfront in the repo's own README, not a hidden catch, and a new Apify account starts with $5 of credit, which covers roughly 60 checks before you'd pay anything.

Who it is for

This is for a business owner or marketer who has no real idea what ChatGPT, Perplexity, Claude or Gemini say about their brand when a prospect asks. It replaces the manual, unreliable version of that job: opening four different AI tools yourself, typing the same question into each one, and trying to remember what came back last time. This does it in one call and gives you a score you can track over time.

Before you start

Install

Step 1: Get the code onto your machine

This downloads the repo and moves into its folder.

git clone https://github.com/khadinakbarlabs/ai-brand-monitor-mcp.git
cd ai-brand-monitor-mcp

Step 2: Install the dependencies

This pulls in the three packages the project needs: the MCP SDK, the Apify client, and a validation library.

npm install

Step 3: Build it

This compiles the TypeScript source into the runnable server in dist/.

npm run build

Step 4: Get your Apify token

Create a free account at Apify, then copy your API token from the Apify console. This is the APIFY_TOKEN the server needs to actually run a check against ChatGPT, Perplexity, Claude and Gemini; the server starts without it, but any real check fails until it's set.

Step 5: Start the server

This starts the MCP server on stdio, ready for an MCP-compatible client like Claude Code or Claude Desktop to connect to it.

node dist/index.js

Run it

The exact command run in the test was node dist/index.js, with no other arguments.

$ node dist/index.js
ai-brand-monitor-mcp running on stdio

That single line to the terminal is correct behaviour for a stdio MCP server: it's now sitting and waiting for a client to talk to it, not crashing and not hanging on an error. Connecting an MCP client and listing its tools turns up exactly two, matching the README word for word:

audit_brand_visibility   run a full GEO audit using standard query templates
check_brand_in_queries   test 1 to 10 of your own custom questions

Troubleshooting

A tool call returns "Missing APIFY_TOKEN." The server starts fine without a token, but every real check needs one. Get a free token from the Apify console and set APIFY_TOKEN in your MCP client's config for this server.

npm run build fails on an older Node version. This repo needs Node 18 or newer. Check your version and upgrade if it's below that.

Nothing prints after node dist/index.js and the terminal just sits there. That's expected, not a bug. It's a stdio server waiting for a client, it only speaks once one connects.

Confused by the two install options in the README. The README leads with a hosted, OAuth, no-install option (Option A) and lists the clone-and-build path second (Option B). This guide tests Option B, the self-hosted path, since that's the one that runs entirely under your own control.

Credit

Built by khadinakbarlabs (https://github.com/khadinakbarlabs/ai-brand-monitor-mcp), MIT license, 28 stars at test time.

If The AI Watchdog just showed you a gap in how AI search talks about your brand, that's the exact audit Focentra runs for clients, so let's get on a call and walk through yours, priced on the call.
This is what Focentra does

Want your real AEO audit?

This is the exact problem Focentra AI solves for clients. Priced on the call, not here.

Talk to Focentra AI
Or learn to build it yourself

Join the 7-Day AI Native Challenge

See the challenge

Download the PDF version

This guide includes affiliate links for tools mentioned in the setup steps. If you sign up through them, I may earn a commission at no extra cost to you.