Design quality gate for AI-generated UI: score any URL on 45 deterministic checks, A–F grade.
Scored across 17 tools
The large family of *_score tools (score, drift, monitor, motion, readiness, tokens, a11y) share a naming shape but each targets a genuinely different artifact (live URL, token file, Lottie file, temporal history), and descriptions include explicit 'When NOT to use' cross-references that steer selection. The weakest area is the cluster of reference-document getters (llms_txt, llms_full_txt, agent_json, catalog, contract, skill_md), which overlap conceptually as 'published documents' though their contents differ. designesy_report also synthesizes score+drift+readiness, creating mild redundancy, but it is clearly framed as the composite capstone.
All names use the designesy_ prefix and snake_case consistently, with the majority following a <subject>_score pattern. The remainder (catalog, contract, compare, report, guardrails, agent_json, skill_md, llms_txt) are noun- or single-verb based rather than verb_noun, which is a minor deviation but still readable and predictable by looking at the list.
At 17 tools the surface is on the heavy side for the domain. Several tools are essentially 'fetch a different published document' (llms_txt, llms_full_txt, agent_json, catalog, skill_md, contract) and could plausibly be consolidated or parameterized, while the core scoring engines legitimately each earn their place.
Coverage is broad across the design-intelligence lifecycle: contract scoring, token/DTCG validation, motion/Lottie validation, drift, continuous monitoring, AI readiness, comparison, guardrail bundle generation, and composite reporting, plus discovery/reference docs. Minor gaps: designesy_a11y_score only returns a PENDING_EXECUTION framework plus a script rather than a score, and there is no per-package detail fetch beyond the catalog list and the contract.
Get the Designesy WCAG 2.2 AA accessibility verification framework: 11 conformance checks (a01-a11) plus a ready-to-run Playwright + axe-core 4.13.0 script template targeting your URL. Use this to audit a site for accessibility violations. When NOT to use: for a full design-contract score (not just a11y), use designesy_score. Does NOT run the scan: axe-core needs a real browser DOM. Returns the 11 checks + a Playwright script you execute locally (npm i -D @axe-core/playwright). The score comes from your local run, not from this tool. Returns JSON: { checks[{id (a01-a11), name, status: "PENDING_EXECUTION"}], playwright_script, install_command, run_command }. Pass config (JSON string) to customize axe.configure(), for example with branding overrides or rule disables. Omit for standard WCAG 2.2 AA.
Inputs: url (required), config, ruleset.
Get the Designesy agent discovery document (/.well-known/agent.json) as reference data: the org identity, authority, discovery endpoints, package index, machine exports, permission policy, contact, and citation templates. Use this when integrating with or enumerating Designesy and need the canonical discovery manifest rather than one specific contract. When NOT to use: for the package list, use designesy_catalog (lighter); for the contract, use designesy_contract. Read-only: no side effects. Returns JSON: { kind: "published_document", source_url, media_type: "application/json", note, omitted_fields[], document }, where document is the published object (schema, name, identity, authority, topics, discovery, ingest, packages, machine_exports, contact, permission, cite, and the rest) with any field written as steps for an AI agent (such as ingest.steps) removed and its path listed in omitted_fields. No parameters.
Inputs: no named parameters.
List the 23 published Designesy packages with versions, URLs, and statuses. Use this to discover what Designesy publishes before fetching a specific contract. When NOT to use: if you already know which package you need, skip this and call designesy_contract directly. Read-only: no side effects. Returns JSON: { package_count, packages[{id, kind, title, version, status, human_url, machine_url}], standing_rules[], machine_exports[] }. No parameters: accepts empty input.
Inputs: no named parameters.
Diff two design systems from live URLs, using the only URL-scoped design-token diff engine. Fetches both URLs in parallel, extracts their :root custom properties, and produces a structured diff across 8 dimensions: tokens added (in A not B), removed (in B not A), renamed (heuristic Levenshtein ≤ 2), value-changed (same name, different value), scale-stop-changed (spacing/radius/color scale steps), contrast-drift-per-pair (WCAG contrast ratio change for shared color tokens), structure-delta (token count + category distribution), and score-delta (runs /score on both URLs and diffs). Use this to answer "what actually changed between two design systems" or "how does our design system differ from a reference". When NOT to use: for single-site drift detection, use designesy_drift_score; for continuous monitoring, use designesy_monitor_score. Executable: fetches both URLs, extracts CSS + tokens, computes diff. No browser needed. Returns JSON: { ok, urlA, urlB, score (0-100, diff completeness), grade, pass, warn, fail, total, tokensA, tokensB, added[], removed[], renamed[], valueChanged[], scaleDiff, structureDelta, contrastDrift[], scoreDelta, checks[] }. Results cached ~24h per URL pair.
Inputs: urlA (required), urlB (required).
Get the Designesy design-system contract: the canonical tokens, motion, acoustic, takt, cadence, typography, components, and verification rules that define what the Designesy org considers legitimate design. Use this when you need the actual contract values (token names and values, motion timings, accessibility rules) to author, check, or bind a design. When NOT to use: for a pass/fail score of a live site, use designesy_score; for an agent-skill-format export, use designesy_skill_md. Read-only; this server caches the fetched contract for 5 minutes. With no arguments it returns the full contract JSON, about 125 KB (roughly 31k tokens). To return less, pass sections, a list of top-level keys such as ["motion", "colors"]: the result holds id, version, and only those keys. section (one name) is the older form and returns { section, data }. An unknown name returns an error that lists every valid key.
Inputs: section, sections.
Get the Designesy Design Review rubric for a qualitative design critique: eight dimensions (Purpose, Clarity, Context, Inclusion, System coherence, Durability, Delight, Responsibility), the output format, and the verification checklist, from the published Design Review kit. Use this when you want a structured rubric to critique a design holistically, rather than a numeric compliance score. When NOT to use: for a deterministic numeric score, use designesy_score; this tool gives you a rubric, not a number. Read-only: it returns reference data and does not evaluate the design. Returns JSON: { kind: "review_rubric", source_url, note, inputs (only when passed), kit { id, title, version, status, purpose, quality_bar, permission }, when_to_use[], required_inputs[], dimensions[8] { num, title, desc }, output_format[], verification_checklist[], anti_patterns[], rationalizations[], kit_prompt_url, kit_prompt_note, omitted_fields[] }. The kit's copy-ready agent prompt is not included; kit_prompt_url is the page where a person can read it. Pass artifact, purpose, context, or rules to have them recorded in inputs (rules defaults to the current contract version).
Inputs: rules, context, purpose, artifact.
Score a live URL for AI-generated UI drift. Its 12 checks detect the four documented 2026 drift failure modes: token fabrication (var() references with no fallback to custom properties declared nowhere: not in a stylesheet, a block or a style attribute), within-session drift (spacing/color/radius value variance), between-session amnesia (inconsistent font stacks, shadows, transitions), and silent breaking changes (z-index chaos, alias chains that end in an undeclared property). Use this when you need to verify whether a site (especially an AI-generated one) is drifting off its own declared token system. When NOT to use: for a full 45-check design-contract score, use designesy_score; for token-file format validation, use designesy_tokens_score. Executable: fetches the URL server-side, extracts all CSS (inline + linked stylesheets), parses custom property declarations (stylesheets, blocks and style attributes) and var() references, runs 12 drift checks. No browser needed. Returns JSON: { ok, url, scope, score (0-100), grade (A-F), pass, warn, fail, skip, total, tokensExtracted, checks[{id, item, category, status, detail}] }. Results cached ~24h per URL.
Inputs: url.
Generate a frozen build-contract bundle for AI coding agents from any design system URL (the product layer). Ingests a site, extracts its :root tokens, and emits 6 outputs: (1) DTCG-format token file, (2) Stylelint config generated from token values, (3) AGENTS.md-format rules with token allowlist, (4) component contract with allowed prop patterns, (5) anti-pattern documentation, (6) DESIGN.md file (Google open spec, google-labs-code/design.md), the de-facto AI-readable design-context standard: YAML front matter plus a markdown body. Use this when you need to turn a design system into the file AI agents read and the lint that enforces it. When NOT to use: for design-contract scoring, use designesy_score; for token-file validation, use designesy_tokens_score; for drift detection, use designesy_drift_score. Executable: fetches the URL, extracts CSS + :root custom properties, generates the bundle. No browser needed. The full bundle grows with the number of tokens a site declares: about 7 KB to about 530 KB of JSON on the 26 sites measured (about 110 KB for designesy.org). To trim it, pass parts, a list of bundle file names such as ["designMd"] or ["tokens", "lintConfig"]: the result keeps score, grade, counts and checks, and its bundle holds only those files, in the order asked. Valid parts: tokens, lintConfig, agentRules, componentContract, antiPatterns, designMd. A single part can still be large on a site with many tokens. An unknown name returns an error that lists the valid parts. Returns JSON: { ok, url, score (0-100, emission completeness), grade, pass, warn, fail, total, tokensExtracted, bundle: { tokens, lintConfig, agentRules, componentContract, antiPatterns, designMd }, checks[{id, item, category, status, detail}] }; with parts it adds parts, and bundle holds only the named files. Results cached ~24h per URL.
Inputs: url, parts.
Get the Designesy /llms-full.txt brief as reference data: authority, discovery endpoints, topics, every published package with its links, standing rules, and anti-patterns. Use this for a fuller picture of the Designesy ecosystem when the short brief from designesy_llms_txt is not enough. When NOT to use: for a quick orientation, use designesy_llms_txt (a few thousand characters; this one runs over ten thousand); for the contract itself, use designesy_contract. Read-only: no side effects. Returns JSON: { kind: "published_document", source_url, media_type: "text/plain", note, omitted_sections[], content }, where content is the published text with any section written as steps or a prompt for an AI agent (such as the ingest protocol and the paste-ready agent prompt) replaced by a one-line marker and named in omitted_sections. No parameters.
Inputs: no named parameters.
Get the Designesy /llms.txt brief as reference data: what Designesy is, canonical links, topics, the published package list, machine exports, standing rules, and contact. Use this first when you don't know what Designesy is; at a few thousand characters it is the cheapest orientation path before pulling heavier artifacts. When NOT to use: for the longer brief, use designesy_llms_full_txt; for the contract itself, use designesy_contract. Read-only: no side effects. Returns JSON: { kind: "published_document", source_url, media_type: "text/plain", note, omitted_sections[], content }, where content is the published text with any section written as steps for an AI agent (such as the ingest steps) replaced by a one-line marker and named in omitted_sections. No parameters.
Inputs: no named parameters.
Score a URL for continuous design-drift governance, the temporal layer over the drift radar. Re-runs the 12 drift checks (d01-d12) on the URL and computes 10 monitor checks (m01-m10): history supplied, last run fresh, drift delta vs baseline, trend slope (3-run trajectory), new violations since last run, resolved since last run (the healing signal), score degradation threshold, token-set mutation, contract version drift, and alert delivered. Alerts come back in the result as alerts[]. No email is sent: email delivery was retired, so m10 is SKIP on every run and is not scored, and the email argument is deprecated and ignored. Pass a history array of prior snapshots (each call's currentSnapshot) to compute deltas; omit it for a first-run baseline. History is checked for shape and bounds (at most 1000 runs, a request of at most 512 KB). Use this to watch a design system over time: "weekly audits at cents per report" (Into Design Systems 2026). When NOT to use: for a single point-in-time drift check, use designesy_drift_score; for design-contract scoring, use designesy_score. Executable: fetches the URL, extracts CSS + :root tokens, runs checks, computes deltas. No browser needed. Returns JSON: { ok, url, score (0-100, governance health; SKIP is not scored), grade (A-F), pass, warn, fail, total, currentSnapshot, baseline, previous, driftChecks, monitorChecks, alerts }. Results are cached ~24h per URL and history: a repeat call with the same URL and history returns the cached run, with m02 measured again.
Inputs: url, email, history.
Validate one Lottie animation file against the Designesy motion contract: its ten checks (m01-m10), drawn from the Lottie spec v1.0.1 and the Designesy §16 Ten Non-Negotiable Motion Standards. Each check returns PASS, FAIL, WARN or SKIP under the contract's own name for its id: m01 the schema's top-level animation object (layer contents are not validated, so a clean file returns SKIP, not PASS), m02 the required fields (v, fr, ip, op, w, h, layers), m03 no deprecated 4.x Bodymovin version, m04 a markers array, m05 a meta object, m06 a reduced-motion path (a marker or slot named for it, or a meta note of an external prefers-reduced-motion wrapper), m07 keyframe easing (linear, ease and ease-in curves fail), m08 duration within 300 ms (longer returns WARN, since a justification cannot be read from a file), m09 layout-property animation and m10 keyboard-initiated motion (both SKIP: they concern the page that embeds the file). Use this to check that a motion asset is well formed and carries reduced-motion support. When NOT to use: for the motion CSS of a whole site, use designesy_score. Executable: fetches the URL or parses the inline JSON; no browser needed. Score: PASS=1, WARN=0.5, FAIL=0, SKIP is not scored; points / scored checks x 100. Returns JSON: { contract_id, contract_version, contract_status, url, lottie_version, layer_count, score (0-100), grade (A-F), pass_count, fail_count, warn_count, skip_count, scoring, checks[{id (m01-m10), name, status (PASS, FAIL, WARN or SKIP), detail}], ten_non_negotiable, provenance, validator_note }. Pass url to fetch a remote Lottie file, or lottie_file to validate an inline JSON string. Provide exactly one.
Inputs: url, lottie_file.
Score a URL for design-system AI readiness, the 6th maturity axis (zeroheight 2026). 10 checks probe the target origin for machine-readable artifacts: DTCG token files, llms.txt, agent.json, MCP endpoint (tools/list), DESIGN.md, token $description, component schemas, sitemap.xml, robots.txt, and Open Graph/Twitter meta. Use this to verify whether a design system is the default context AI tools build from, or whether AI is silently working around it. When NOT to use: for full design-contract scoring, use designesy_score; for AI-drift detection, use designesy_drift_score. Executable: fetches the URL and probes the origin via HEAD/GET for each artifact. No browser needed. Returns JSON: { ok, url, score (0-100), grade (A-F), pass, warn, fail, total, checks[{id, item, category, status, detail}] }. Results cached ~24h per URL.
Inputs: url.
Generate a unified design-intelligence report for a single URL, the synthesis capstone. Fires /score (45-check audit), /drift (12-check drift radar), and /readiness (10-check AI readiness) in parallel, then computes a weighted composite: score × 0.5 + drift × 0.3 + readiness × 0.2. One input, one output, one composite grade. Use this when you need a single holistic assessment instead of three separate scans, or when sharing a design-intelligence verdict (the report is the most shareable surface). When NOT to use: for just the audit score, use designesy_score; for just drift, use designesy_drift_score; for just AI readiness, use designesy_readiness_score. Executable: fires 3 internal APIs in parallel, each fetches the target URL. No browser needed. The full result is large: about 75 to 100 KB of JSON on the ten sites measured, on the order of 20,000 tokens or more, because it carries each engine's whole result beside the merged check list. To trim it, pass detail "summary": the same keys with fewer rows, about 4 to 31 KB on the same sites. It keeps the composite, the totals, each engine's score, grade and counts, and only the checks and synthesis entries that did not PASS; omitted counts what was left out. detail "full" (the default) returns everything. Returns JSON: { ok, url, compositeScore (0-100), compositeGrade (A-F), score { sub-result }, drift { sub-result }, readiness { sub-result }, totalChecks, totalPass, totalWarn, totalFail, totalSkip, totalManual, checks[] (all checks across all engines, tagged with engine), synthesis[] (8 synthesis checks verifying the report ran correctly), appUrl (standalone interactive dashboard URL) }; with detail "summary" it adds detail and omitted. Results cached ~24h per URL. MCP Apps: hosts that support io.modelcontextprotocol/ui render an interactive dashboard inline; others get the JSON plus an appUrl link.
Inputs: url (required), detail.
Score a live URL against the Designesy design contract: a deterministic 45-check verification engine that returns a numeric score, letter grade (A-F), and per-check breakdown. Use this to audit whether a website or AI-generated UI complies with a real design contract (tokens, motion, accessibility, cadence, takt, typography, copywriting). When NOT to use: for token-file validation only, use designesy_tokens_score; for a Lottie file, use designesy_motion_score; for a qualitative critique, use designesy_design_review. Executable: fetches the URL server-side, extracts CSS, runs 45 checks. Results cached ~24h per URL. Checks needing a live browser (Core Web Vitals, sound toggle, overflow) return MANUAL (not FAIL); run the full audit (/api/score/audit) to resolve them. Checks that are not applicable to the site (no tokens, no buttons, no DESIGN.md) return SKIP (N/A). The default format returns the engine's JSON: { score (0-100), grade (A-F), pass, warn, fail, skip, manual, total, checks[{id, item, category, status, detail, remediation, weight}], categoryScores, contractVersion, receipt }. Pass format="canonical" for review-findings.json schema, "review" for markdown, or "google" for design.md-compatible output. Pass scope="contract" or "universal" to set the scoring scope instead of auto-detecting it.
Inputs: url, scope, format.
Get the Designesy SKILL.md: the agent-skill-format export of the design-system contract, for a user to save into .agents/skills/ or a system prompt so a coding agent builds UI to the contract (tokens, anti-patterns, rules, verification). Use this when the user wants the contract in that form. When NOT to use: for the raw contract JSON, use designesy_contract; for scoring, use designesy_score. Read-only: no side effects. Returns JSON: { kind: "published_document", source_url, media_type: "text/markdown", note, omitted_sections[], content }, where content is the SKILL.md markdown (tens of thousands of characters) with any section written as steps or a prompt for an AI agent replaced by a one-line marker and named in omitted_sections. No parameters.
Inputs: no named parameters.
Validate a design token file against the W3C Design Tokens Community Group (DTCG) 2025.10 Final Community Group Report (the spec's first stable version, published Oct 28 2025 as a Candidate Recommendation and considered stable). Returns 10 conformance checks (t01-t10) with PASS, FAIL, WARN or SKIP. Use this to verify a tokens.json (or any DTCG token export) is structurally correct: $type (direct or inherited) present, one $value or $ref per token (a JSON Pointer $ref must resolve), structured colors (colorSpace + components rather than bare hex), dimensions in the 2025.10 { value, unit } form, and a $schema pointer to designtokens.org (a Designesy policy: DTCG does not require $schema). With 84% of teams now using design tokens (zeroheight Design Systems Report 2025, up from 56% in 2024) and the spec finally stable, every adopting team needs a validator. When NOT to use: for scoring a whole live site (not just its token file), use designesy_score. Executable: fetches the URL or parses the raw JSON you provide, runs 10 checks server-side. No browser needed. Score: PASS=1, WARN=0.5, FAIL=0, SKIP is not scored; points / scored checks x 100. Returns JSON: { contract_id, contract_version, contract_status, url, total_tokens, score (0-100), grade (A-F), pass_count, fail_count, warn_count, skip_count, scoring, checks[{id (t01-t10), name, status (PASS, FAIL, WARN or SKIP), detail}], provenance, validator_note }. Pass url to fetch a remote token file, or dtcg_file to validate an inline JSON string. Provide exactly one.
Inputs: url, dtcg_file.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Score a public website 0 to 100 for the design habits that make it look AI-made.
Anti-slop design taste for AI coding agents: art directions, section code, 0-100 page critic.
Checks, prompts and skills for designers working with AI: contrast, a11y, heuristics, critique.
A design harness for AI coding agents: design systems, design principles, icons and UI review.
Add one secure layer between your agents and this server.
