The tools your agent can call

What each of Rabbot's fifteen MCP tools answers, the two that can change anything, and why an agent can preview a fix but never apply one.

Updated 13 October 2026

Rabbot’s MCP server holds fifteen tools bound to one store at a time. Thirteen only read, and two of those read Rabbot’s own documentation rather than your store. Two write, and each refuses on anything that would matter if it went wrong. No tool applies a fix: that takes your two taps in the Rabbot app.

The thirteen read tools

  • get_protection_status returns your plan, how many monitored URLs you are using against your cap, when the last full check ran, and how many open issues sit at each severity.
  • whats_blocking_sales returns what is stopping a sale right now, grouped into product stories with a chip for the channel involved, rather than a flat list of rows.
  • list_open_issues returns every open problem, most severe first, with the page, the rule it tripped, the severity, the detail, and when it opened.
  • get_url_timeline returns one page’s history, newest first: changes, when an issue opened, when it was fixed, and clean runs in between. Seven days back on the free plan, the full history on paid plans.
  • list_recent_catches returns what Rabbot caught and confirmed fixed in the last 30 days. Use it when someone asks what Rabbot has actually caught.
  • get_url_health returns one monitored URL’s last check time, what that check found, and its open issues.
  • get_store_check_detail returns one check in full: what Rabbot looked at, what it found, and when.
  • list_day0_findings returns the audit Rabbot ran the day it was installed: price, availability, status, and completeness findings from your feed at that point.
  • notifications_sent returns what actually went out in the last seven days, by channel, straight from the send log, so you can check an alert really was sent before you go looking for it elsewhere.
  • fix_preview returns what a fix on a finding would write: the field, the value it has now and the value Rabbot would put back. It reads only.
  • fix_apply never applies a fix. It answers with the link to the finding in the Rabbot app, in your Shopify admin, where you review the values and confirm in two taps. An agent cannot skip that step.
  • search_rabbot_docs searches Rabbot’s own documentation, the pages under rabbot.ai/docs, so an agent can answer how-do-I questions about Rabbot from the docs instead of guessing. It reads nothing from your store.
  • get_rabbot_doc returns one page of Rabbot’s own documentation in full. It reads nothing from your store.

The two write tools

  • dismiss_finding marks a finding as intentional, for one page or store-wide, so Rabbot stops flagging it. It refuses on anything blocking a sale, with two documented exceptions that can be intentional: a deliberate password gate or a custom purchase flow. Everything else in that category cannot be silenced by an agent, only fixed. Dismissing a finding mutes the alert. It does not fake a fix. If the underlying problem is later repaired, the finding closes and the dismissal goes with it. If it comes back after that, Rabbot alerts again as normal.
  • fix_undo reverses a fix Rabbot made for you, within the 30 days the undo stays open. It refuses once the window has closed, and it refuses an undo that would send visitors in a circle. The one fix with no undo is a redirect chain pointed straight at its end, and the card says so before you tap.

Asking for these in plain language

You do not need to know the tool names to use them. Ask your agent a plain question, “what’s blocking sales” or “show me this page’s history”, and it picks the right tool on its own. The names above are here so you know what exists and what each one is bound to answer, not because you need to type them yourself.

Still stuck? Email support@rabbot.ai and a real person reads it.