Skip to content
Rubric

UX expertise your agent can call.

Rubric serves a curated UX corpus to coding agents over MCP. Your agent designs against what actually works, then audits a live site against the same rules, with findings, severity and the fix.

Connect 
claude mcp add --transport http rubric https://rubric.design/mcp
Or install the plugin — it connects for you 
/plugin marketplace add things-studio/rubric-plugins
/plugin install rubric-audit@rubric

A browser opens, you approve, and that is the whole setup — no key to paste. Any MCP client works, because the server is plain streamable HTTP. Where no browser can open — CI, a cron job — create a key and send it as Authorization: Bearer … instead.

Published units
554
Findings
617
Surfaces
17
Disciplines
6

A model knows UX in general. It does not know your checkout.

Ask an agent for design advice and you get the average of everything written about design. Rubric replaces that with a specific corpus: every unit is one rule, judged once for severity, with how to find it, what it costs, and what to do. The same units serve design time and audit time, so the thing that builds it and the thing that checks it are reading the same page.

Before you build

Prescriptive guidance, not a description of a problem. “Give every focusable element a visible focus indicator”, with the mechanism that makes it work and what breaks without it.

  • search_ux_knowledge
  • get_guidelines

After you ship

A checklist your agent evaluates against the real page, then findings ranked by severity and a report you can hand to whoever fixes it. Nothing is scored or averaged. Severity is the only ranking there is.

  • get_checklist
  • get_findings
  • create_audit
  • submit_findings
  • get_report

Plugins

The server holds the knowledge. It does not hold the method — the order of the calls, what counts as evidence, and why a check you skipped has to be reported rather than omitted. That travels in a Claude Code plugin, and it installs the MCP connection with it.

rubric-audit

UX audit and reporting

Walk a live site, judge it against the checklist, and produce a report with severity, evidence and the fix. Asks for your API key once and wires the server up for you. 

/plugin marketplace add things-studio/rubric-plugins
/plugin install rubric-audit@rubric
things-studio/rubric-plugins

A design-time plugin, for pulling guidelines before you build, is next.

What it looks like in the chat

An illustration, not a recording. The tool calls and unit keys are real.

Designing to the corpus

The guidelines arrive before the code does, so the thing that gets built is the thing an audit would have asked for.

agent · new-store

Auditing what is already live

The same units, used the other way round: evaluated against a real page, then ranked by severity and written up.

agent · acme.com

What comes back

A finding is four separate blocks, never one paragraph. That is also where the free tier ends: the issue and its severity are free on every call, and the rest is what you pay for. Below is a real one, from checkout.guest-option.

critical checkout.guest-option Guest checkout available without account creation
Issue free
Checkout forces users to create an account before paying; there is no clear guest option.
Impact paid
Forced-account checkout asks the shopper to complete a second task — creating and remembering credentials — before they can finish the one they came for, at the exact point where they expected the transaction to be almost done.
Actions paid
  • · Add a “Continue as guest” option at the top of the checkout sign-in step, equally prominent as “Sign in” and “Create account”.
  • · Offer one-click account creation on the order-confirmation page, pre-filling all data from the order.
  • · Never reveal an “account already exists” error before the user enters a password — protects against email enumeration.
References paid
Apple Store · ASOS · Shopify default checkout

Works with

Rubric is a standard MCP server over streamable HTTP, authenticated with a header. If your client speaks MCP, it works. There is nothing Rubric-specific to install.

Pricing

Money on an account, and each call takes its price out. No seats, no monthly minimum, nothing to cancel.

Free

For trying it on something real.

  • The full checklist, every item
  • Every finding’s issue and severity
  • 100 calls a day
  • No card
Sign up

Pay as you go

usage-based

For work you are shipping.

  • Everything in Free
  • Impact, actions and references on every finding
  • No daily cap
  • Add funds, spend them down, top up when you want
Sign up

Enterprise

For a team with a procurement process.

  • Invoicing and volume terms
  • Corpus extensions for your own patterns
  • A person who answers
Contact us