DS4AI MCP
# DS4AI MCP server
The aggregate endpoint of **DS4AI, the Design Suite for AI, from [Polymathie-Studio](https://github.com/Polymathie-Studio)**. It serves the suite as callable MCP tools, so a person working with an AI reaches the conformance auditor and the generators through their agent, without calling a function. Unlike the `@proof-of-coord` standards servers, which point at canonical source, this server *runs* the primitives, and because it is a process rather than a browser it can fetch a URL and audit the served HTML.
## Use it
This server is hosted as a public HTTP endpoint, and is published to npm and JSR as `@polymathie/ds4ai`. Connect the hosted server with nothing to install:
```json
{
"mcpServers": {
"ds4ai": { "type": "http", "url": "https://ds4ai-production.up.railway.app/mcp" }
}
}
```
Or run it locally, no clone:
```
npx @polymathie/ds4ai # from npm
deno run -A jsr:@polymathie/ds4ai # from JSR (Deno)
```
Add it to an MCP client (Claude Code, Cursor, and the rest) either as the hosted HTTP server above, or as a stdio server whose command is `npx @polymathie/ds4ai`. The four MCP servers are listed at [integritysuite.org/built](https://integritysuite.org/built), and the full reference for every connection method is [Using the MCP servers](https://github.com/durgadasji/standards-index/blob/main/using-the-mcp-servers.md).
## Tools
- **audit_surface** `{ url? , html? }`: audit a shipped surface across all six invisible-correctness axes with the MISSING conformance auditor. Give a URL and it fetches and audits the served HTML. Declares the axes a static check cannot judge rather than reporting them clean.
- **generate_head**, **generate_jsonld**, **generate_sitemap**, **generate_robots**, **audit_findability** (BEACON): findability head tags, JSON-LD, sitemap, robots.txt, and the depth findability audit.
- **generate_image_markup**, **generate_cache_config**, **audit_delivery** (FLEET): responsive image and picture markup, cache-header config per host, and the depth delivery audit.
- **solve_palette**, **check_contrast** (TEMPER): solve a full palette to contrast floors for a mode, and check the WCAG contrast between two colors.
Every result carries a provenance stamp declaring the server and bundled primitive versions.
## Resources
- `ds4ai://manifest`: the DS4AI family manifest (read generically, so it tracks the manifest as its data source).
- `ds4ai://agents`: the agent-instruction (AGENTS.md), so a connected agent gets the compose-the-suite instruction for free.
- `ds4ai://standard`: the MISSING standard.
## How it is built
TypeScript on the MCP SDK with zod schemas, provenance-stamped, following the `@proof-of-coord` pattern. The zero-dependency primitives (TEMPER, BEACON, FLEET, and the MISSING conformance auditor) are bundled in, so the server is self-contained, and each tool stamps the primitive version it bundled. Published to npm and JSR under `@polymathie`.
## Security
The URL-auditing path (`audit_surface` fetching a live URL) is built to a defensive threat model and is adversarially tested and hardened before release against server-side request forgery, denial-of-service, and resource-exhaustion attacks. Fetched content is treated as untrusted input: it is parsed, never executed, and the fetcher connects only to validated public addresses. The implementation is open for review under Apache-2.0.
## Part of DS4AI
DS4AI, the Design Suite for AI, from Polymathie-Studio, joins PT4AI and PC4AI in the family of corpora served to an AI. The instruments and the MISSING standard live at [Polymathie-Studio](https://github.com/Polymathie-Studio).
## License
Apache-2.0 for the server. Copyright 2026 Regis Lloyd Chapman. See `LICENSE` and `NOTICE`.
The bundled resources keep their own licenses, from the [Polymathie-Studio/missing](https://github.com/Polymathie-Studio/missing) repository: `resources/standard.md` (the MISSING standard) is dual-licensed CC-BY-4.0; the manifest and `AGENTS.md` are Apache-2.0. The vendored primitive source under `src/vendor/` is Apache-2.0, from its instrument repository.
TDQS
Scored across 11 tools
Each tool names a distinct artifact or audit focus (head, JSON-LD, sitemap, robots, images, cache, palette, contrast). The only mild overlap is audit_surface versus audit_findability/audit_delivery, but the descriptions clearly differentiate a broad holistic audit from in-depth single-axis audits.
Tool names consistently follow a verb_noun pattern: generate_* for output serializers, audit_* for auditors, plus solve_palette and check_contrast. There is no mixed casing or vague verb style, making the naming pattern predictable and readable.
11 tools is well within the ideal scope for a web-correctness toolkit. Each generator and auditor addresses a distinct deliverable or audit pass, and none feels redundant or out of place.
The set covers core generation (head, structured data, sitemap, robots, images, cache) and auditing (surface, findability, delivery) plus color accessibility. The main gap is that index/noindex control is mentioned as living in the head, but no tool directly emits that meta robots directive.