Rubric
UX guidelines, research and audit checklists for coding agents, plus your own Rubric audit reports and analytics. Remote server with OAuth: claude mcp add --transport http rubric https://rubric.design/mcp
Размещённый MCP-сервер
npx add-mcp 'https://rubric.design/mcp'Устанавливается в Claude Code, Codex, Cursor и другие
Документация
Rubric
Curated UX knowledge, served to coding agents over MCP.
Rubric answers two questions: what should I build (design-time guidelines) and what is wrong with this page (an audit checklist an agent runs, then submits its observations against). Findings carry severity, impact, concrete actions and citations. There are no scores — severity is the only ranking signal, judged once per unit rather than computed per site.
Connect
Endpoint: https://rubric.design/mcp — streamable HTTP, nothing to install.
Two ways in:
- OAuth 2.1. Add the endpoint as a custom connector and sign in. No key is
pasted anywhere. Discovery starts at a 401 carrying
WWW-Authenticate, which names https://rubric.design/.well-known/oauth-protected-resource/mcp. - An API key, for CI, a server-side script, or stdio — anywhere no human can
open a browser to approve. Send it as
Authorization: Bearer <key>. Keys are made at https://rubric.design/app/keys.
There is no unauthenticated access and no shared demo key.
Tools
| tool | what it does | free tier returns |
|---|---|---|
search_ux_knowledge | Find units by query and facets. | title, guideline |
get_guidelines | Design-time prescriptions for a surface or discipline. | guideline |
get_checklist | What to check on a page, with navigation hints. | everything |
get_findings | The written finding for a unit and outcome. | issue, severity |
create_audit | Start an audit of a URL. | — |
submit_findings | Record what the agent judged, unit by unit. | — |
get_report | Read an audit back, as JSON or markdown. | issue, severity |
A paid call adds the impact, the actions and the references.
Limits
Free calls are capped per day, per account, by call count. A paid plan includes a monthly allowance of MCP calls. An audit from your own agent uses no runs: runs are spent only by managed runs on Rubric's browser. Plans are at https://rubric.design/app/plan. Connecting an account grants nothing on its own.
Machine-readable
- MCP server card: https://rubric.design/.well-known/mcp/server-card.json
- API catalog: https://rubric.design/.well-known/api-catalog
- Protected resource metadata: https://rubric.design/.well-known/oauth-protected-resource/mcp
- Agent skills: https://rubric.design/.well-known/agent-skills/index.json
- Health: https://rubric.design/health