Rabbot’s MCP server: ask your agent about your store.
Rabbot runs its own MCP server. Thirteen tools only read: eleven read a store's status and two read Rabbot's own documentation. Two narrow, guarded tools write: one marks a finding as intentional so Rabbot stops flagging it, and one undoes a fix Rabbot made. An agent can preview a fix and can never apply one. Any MCP-capable agent can use it once you mint a token.
Fifteen tools, plainly named.
Every tool binds to one authenticated store per call. Thirteen of them read. Two of them write, and only in a narrow, guarded way.
get_protection_status whats_blocking_sales list_open_issues get_url_timeline list_recent_catches get_url_health get_store_check_detail list_day0_findings notifications_sent fix_preview fix_apply search_rabbot_docs get_rabbot_doc dismiss_finding write, guarded fix_undo write, guarded
Rabbot never reports a clean bill of health it can't back up. If a Critical issue is open
anywhere on your store, whats_blocking_sales says so, even when nothing it
classes as sale-blocking is open.
A raw integration hands back rows. Rabbot hands back a verdict.
Point a generic tool at a store's own data and it returns exactly what you asked for: a list, a count, a field. Nothing more.
A raw integration
- Returns rows: a list of products, a list of orders, a list of pages.
- The judgment is your job. Deciding whether a row is a problem is on you.
- No memory. It only knows what it can see right now.
Rabbot's MCP
whats_blocking_salesnames the exact problem, already triaged by severity.get_url_timelinehands back a page’s history, not just today’s snapshot: last 7 days free, full history on paid plans.- History Rabbot already recorded. An integration that starts watching today cannot rebuild it.
Three steps, about a minute.
- Open Settings in your installed app and find Agent access (MCP).
- Mint a token. It is shown once, so copy it now.
- Paste the command below, or drop the config into your agent’s client.
claude mcp add --transport http rabbot https://app.rabbot.ai/mcp --header "Authorization: Bearer YOUR_TOKEN"Settings builds this command with your real token already in it. Paste it as-is.
{
"mcpServers": {
"rabbot": {
"type": "http",
"url": "https://app.rabbot.ai/mcp",
"headers": {
"Authorization": "Bearer YOUR_TOKEN"
}
}
}
}Works with any MCP-capable agent. Revoke a token in Settings and access ends.
Rabbot answers inside Shopify Sidekick too.
Sidekick is the assistant built into the Shopify admin, and it can call six read-only Rabbot tools. Every answer describes your last completed check, and Sidekick is never sent channel verdicts or order data. No token is involved: install Rabbot, then approve it the first time Sidekick asks.
One token, scoped to one store.
- Rabbot stores only the token’s hash. The raw token is shown once, at mint, and never logged.
- Minting a new token retires the old one. Re-minting is how you rotate it.
- Five requests a second per token, ten in a burst. A flood of bad tokens gets capped before it reaches the database.
- This token is separate from how a merchant signs in. Revoking it never touches the store’s Shopify connection.
Questions worth a straight answer.
Does Rabbot have an MCP server?
Yes. Rabbot ships its own MCP server: fifteen tools bound to one store at a time, mounted in the same server that runs the app.
Is the MCP server free?
Yes. Agent access is free on every plan, including the free tier. Mint a token from Settings once Rabbot is installed.
Can an agent change my store through it?
Two tools write. dismiss_finding marks a finding as intentional so Rabbot stops flagging it. It refuses anything blocking a sale, with two documented exceptions that can be intentional: a deliberate password gate or a custom purchase flow. fix_undo reverses a fix Rabbot made for you. An agent can see what a fix would write with fix_preview, and it can never apply one. fix_apply answers with the link to the finding, where you confirm in two taps in the Rabbot app. Every other tool only reads.
Does an agent need my Shopify login?
No. It needs a bearer token you mint from Settings, scoped to one store and revocable at any time. That token shares no code path with how you sign in.
Does it work with agents other than Claude?
Yes. Any MCP-capable agent that can send a bearer token in a header can connect to it.
Give your agent the same watch you already trust.
Mint a token from Settings and connect the agent you already use.
Free forever · no card required · an agent previews a fix, never applies one