Connect Your Heatmaps to Claude via MCP (1 Minute, One Command)

TL;DR

HeatMapX speaks MCP (the open standard for plugging tools into AI assistants). Free on every plan

Setup is two steps — issue an API key, run one command — about a minute

Once connected, Claude can query your heatmaps, page stats and A/B test results inside the conversation

"Open the dashboard → copy the numbers → paste them into the AI" — that round-trip gets old fast. With MCP, Claude fetches your HeatMapX data itself, and the whole analysis happens in one conversation.

What is MCP?

MCP (Model Context Protocol) is the standard for connecting external tools to AI assistants like Claude. Think of it as a universal power socket: plug in once, and the AI grabs the data it needs, whenever it needs it.

Connect from claude.ai / Claude Desktop (easiest — 3 clicks)

On claude.ai or Claude Desktop you don't need a terminal or an API key:

Open Settings → Connectors → "Add custom connector"
Enter "HeatMapX" as the name and https://heatmapx.com/api/mcp as the URL → add → "Connect"
Log in to HeatMapX and review the consent screen (read access / create & start / stop experiments / register sites), then click "Allow"

That's it — ask away in chat. You can disconnect anytime in HeatMapX Settings → Connected apps.

Connect from Claude Code (terminal, 2 steps)

1. Issue an API key

In your HeatMapX dashboard, go to Settings → API keys and click "Issue new key". The key is shown only once — copy it right away.

2. Run one command

In your terminal (replace YOUR_API_KEY with the key you copied):

claude mcp add --transport http heatmapx https://heatmapx.com/api/mcp --header "Authorization: Bearer YOUR_API_KEY"

Done. Restart Claude Code (or start a new conversation) and the HeatMapX tools are available.

What you can do

Once connected, questions like these just work:

"Show me my sites" → your registered sites and URLs
"Which page on kpop-ticket.com has the worst scroll depth?" → found from per-page stats
"Summarize the heatmap for /pricing" → click hotspots and scroll reach
"How's my A/B test doing? Can I call a winner yet?" → variant metrics plus a statistically grounded winner suggestion

HeatMapX supplies the numbers; Claude does the interpreting. Each side doing what it's best at.

Not just reading: create and launch A/B tests

Connect with write access and the whole loop happens in conversation:

"Create an A/B test on my homepage: two CTA copies" → a draft is created
"Start it" → Claude summarizes the experiment and always asks for your confirmation → say OK and it goes live
"Stop it" → same confirmation, then it stops

Safety by design:

Creation always lands as a draft (nothing serves traffic on its own)
Start / stop never execute without your explicit approval (two-step confirmation)
Deleting is not possible via AI (irreversible operations are off the table)
Permissions are spelled out on the consent screen and revocable anytime in Settings → Connected apps

Available tools

Tool	What it returns
list_sites	Your registered sites
get_page_stats	Per-page sessions, clicks, average scroll depth
get_heatmap_summary	Heatmap aggregates for one page (with time window)
list_experiments	Your A/B experiments
get_experiment_results	Variant metrics, uplift, winner suggestion

Five read tools plus five write tools (create_site / create_experiment / start·pause·stop_experiment). Write tools work only on connections where you explicitly approved write access, and start/stop always require an in-conversation confirmation.

If it doesn't connect — checklist

Key pasted wrong? Make sure you pasted the whole string starting with hmx_live_
Key revoked? Check Settings → API keys for an active key

Pricing

MCP access is free on every plan (with a sensible burst limit only). Heatmaps and A/B tests, straight from Claude — give it a try.