/// getting-started

Building with AI

PDS is designed to be read by machines as well as people. Every component, token, and rule is published in formats an AI coding assistant can ingest — so a model building on PDS uses the real contract instead of guessing.

Machine-readable docs

Two site-root files give any LLM a complete, current picture of the system in a single fetch:

llms.txt

/llms.txt

A compact index — every component and page with its description and URL. The fast map an assistant reads first to orient itself.

llms-full.txt

/llms-full.txt

The complete corpus — full prose, props, and guidelines inlined. One file, the whole system, for models that want everything in context.

The /ai endpoints

For structured, per-component access, the site exposes a small JSON + Markdown API. It mirrors exactly what you read on the component pages — same source, same prose — so an assistant and a human never disagree about the contract.

Endpoint	Returns
`/ai/registry.json`	The machine index — every component keyed by `category/slug`, with description, status, and import names.
`/ai/<category>/<slug>.md`	One component as clean Markdown — e.g. `/ai/inputs/button.md`. Props, examples, and guidance, ready to paste into a prompt.

Every component page also carries a Copy-for-LLM button that copies that page’s Markdown to your clipboard — the quickest way to hand one component’s exact contract to an assistant mid-task.

The MCP server

For assistants that speak the Model Context Protocol, @protocore/pds-mcp exposes the whole system as tools — no copy-pasting, no stale context. Register it with one command:

add the PDS MCP server

claude mcp add pds -- npx -y @protocore/pds-mcp

Once registered, the assistant has five tools:

Tool	Does
`list_components`	Enumerate every component with its category, one-line description, and status.
`get_component`	Fetch one component's full doc — props, examples, do/don't, and a11y notes.
`get_tokens`	Return the token layer — colours, spacing, type scale, and motion — as structured data.
`get_guidelines`	Return the system doctrine — the differentiation contract and styling rules.
`search_components`	Find components by name, description, or use case for a natural-language query.

Always current

The MCP server serves the same registry that builds this site, so an assistant reads the components exactly as they ship — not a snapshot from its training data.

Authorization

Machine-readable ≠ freely usable

These endpoints exist to enable AI-assisted development on projects authorized by Protocore. Their existence does not grant a licence: PDS is proprietary, and building with it — by hand or with an assistant — requires prior written authorization from Protocore. Automated access does not waive that requirement, and ingestion into datasets or model-training corpora is expressly not permitted.

Read the full terms on the License & IP page, or request authorization at contact@protocore.io.