—

For agents

aindf — mcp
$ mcp connect aindf://
aindf://mcp
4 tools over the whole contract graph · spec v0.1
handshake ok
Entry points
0.llms.txtindex 1.list-by-facetenumerate 2.slot-acceptscontract 3.applicable-modifiersgate 4.get-presetresolve
Sample · request
{
  "tool": "applicable-modifiers",
  "args": {
    "component": "Button",
    "context":   { "interaction": "button" }
  }
}
Sample · response
{
  "allowed":  ["surface:accent", "size:sm"],
  "excluded": ["effect:loading"],
  "reason":   "interaction=button gates effect:loading off"
}
$ curl aindf.oleg.design/llms.txt ↗

The whole point: a conformant system answers. An agent does not scrape a screenshot or guess from prose — it queries the contract graph and gets typed, enumerable answers. Each tool reads one schema.

  1. [0]
    llms.txt

    Flat index of the whole site and every artifact. Crawl-friendly — no parsing of nav or sitemap required.

    ↗
  2. [1]
    tool · list-by-facet

    Enumerate components by axis: layer · role · renderTarget · interaction. Returns ids with metadata. Reads taxonomy.

    ↗
  3. [2]
    tool · slot-accepts

    Given a slot, return its accepts list, cardinality and order. Composition becomes a type problem. Reads slots.

    ↗
  4. [3]
    tool · applicable-modifiers

    Given a component + context, return legal modifiers and excluded ones with a reason — the when{} gate, made queryable. Reads applicability.

    ↗
  5. [4]
    tool · get-preset

    Resolve a preset name to a typed composition tree with every slot filled. Reads presets (and patterns for parametrized recipes).

    ↗
  6. [+]
    conformance probe

    Point the validator at a DS's declared coverage; receive a verified pass / fail report against spec §8. Partial coverage still earns a badge.

    →

A design system is fully ai-native when it ships all six principles from one source. Most ship a subset. aindf is the spec for the whole graph.

$ curl https://aindf.oleg.design/llms.txt