AI TOOLS · MCP
Ask your coding tool.
It reads your storefront.
Seven read-only tools give your assistant the same evidence as the dashboard, next to your code: what costs the most, what broke in this file, what a shopper did, and whether the fix worked.
From “what costs the most” to “did the fix work”.
An illustrative session in a theme repository. Every call is read-only and scoped to one store.
CLAUDE CODE · THEME REPOSITORY
Your assistant, with your storefront in context.
What’s breaking in assets/cart.js, and did yesterday’s fix work?
list_problemsfile: "assets/cart.js" → 2 open problems, ranked by impact
get_problemCart update fails after a delivery change · cart.ts:184 · theme · 486 sessions
verify_fixrelease storefront@2.18.5 → 0 occurrences in 26 h vs 41 in the 26 h before · stopped
ASSISTANTTwo problems touch cart.js. The delivery-change failure stopped after 2.18.5: no occurrences in 26 hours against 41 before, with cart traffic present. The second one is still open; want me to look at it?
Connect in a minute.
Create a read-only key under Settings → AI agents and add the Dozenfold MCP server to Claude Code, Cursor or another MCP client. The key reads one store and cannot change anything. MCP access is included in every plan.
See the evidence it readsAsk about the code in front of you.
list_problems ranks open problems by estimated impact and filters by file, funnel stage, release or kind, so an assistant in a theme repo can ask about assets/cart.js. get_problem returns the full incident: mapped stack, linked causes, shopper steps, environment, owner and replay sessions.
Explore an issueCheck the result after shipping.
compare_releases shows new, regressed and spiking problems per release. verify_fix compares a problem’s occurrences after its fix release with the same length of time before, and reports too_early instead of guessing.
Follow a releaseA LITTLE MORE DETAIL
Good questions.
Useful answers.
Which tools are available?
get_store_health, list_problems, get_problem, list_problem_sessions, get_session_timeline, compare_releases and verify_fix. All are read-only.
Which clients work?
Claude Code, Cursor and any client that speaks MCP over streamable HTTP. Setup is one endpoint, https://app.dozenfold.com/mcp, and one Bearer key; Claude Code takes a single claude mcp add command.
Does my assistant see shopper personal data?
It sees the same anonymous evidence as the dashboard: no input values, no identities, no request bodies.
Is this the AI reliability engineer on your roadmap?
No. MCP gives your own tools access today. Dozenfold’s own agent that investigates, works on fixes and follows outcomes is planned, within permissions you set.
CONNECTED PRODUCT WORKFLOWS
Follow the next connection.
An issue, a journey, a page or a release. Each view gives you a way into the next.SEE IT ON YOUR STORE
Bring a real problem to the demo.
Bring the journey you want to improve and the person who ships your changes. We’ll look at your first evidence together.
Book a demo Estimate your price and contact us · Guided setup