Agent surfaces

The surfaces that make this store usable by AI agents

Two kinds of surface work together: instruction files that tell an agent how to use the catalog, and an MCP server that lets it actually run the fitment tools. Open any of them below.

Instruction files — “how to use this store”

agents.md

View raw →

The primary instruction guide: what the store is, the tools available, and the step-by-step fitment flow (resolve fit type → Y/M/M → body → roof type → find racks).

Read by: ChatGPT, Claude, Copilot, and crawlers that follow the agents.md convention

Note: This is exactly the file Shopify auto-serves and lets merchants customize via templates/agents.md.liquid.

llms.txt

View raw →

A concise pointer file — a short summary plus links to agents.md, the full instructions, and the MCP server.

Read by: Agents looking for a quick map of the site

Note: Follows the emerging llms.txt convention.

llms-full.txt

View raw →

The full agent guide plus a live appendix of the catalog's categories, generated at request time.

Read by: Agents that want everything in one fetch

Note: Same content as agents.md, expanded with real catalog data.

MCP server — “do the work”

Model Context Protocol endpoint

A public server exposing the 10 fitment tools. This is a protocol endpoint, not a web page — opening it in a browser shows a JSON-RPC error, which is expected. Add it as a Connectorin ChatGPT or Claude, then ask “what fits my vehicle?”

https://rackattack.teifi.dev/api/mcp

Transport: streamable HTTP · no auth · tools: search_products, get_fit_types, get_vehicle_makes/models/bodies, get_fitment_options, find_fitting_racks, add_to_cart, …

How agentic discovery works →View the deck →