Agentic discovery for
fitment-heavy catalogs
Enabling Shopify sellers in automotive & parts to be found, understood, and transacted by AI agents — ChatGPT, Claude, Copilot.
Buyers are starting to shop through agents
A growing share of product research — and increasingly the purchase itself — happens inside an AI assistant, not a search bar.
For parts & automotive, the first question is always the same
“Will this fit my vehicle?” The agent has to answer that before it can recommend anything.
If the agent can’t resolve fitment…
…it recommends a competitor, hedges (“check the manufacturer”), or invents an answer. The sale leaks.
Fitment is uniquely hard for AI agents
Buyer side
“Will this fit my truck?”
Said in plain language. No Year / Make / Model form. The agent must extract the vehicle and know what’s still missing.
Merchant side
One SKU fits hundreds of vehicles
…but only in specific configurations. The right answer depends on body style and roof type — not just Y/M/M.
Example: a 2020 Toyota Tacoma returns four different roof-rack systems depending on whether it has a naked roof, raised rails, a track, or fixed mounting points.
How an AI agent actually “sees” a product
months stale · approximate
live but unstructured · scraped HTML
structured · queryable · authoritative
The first two are guesses. Fitment can’t be guessed — it has to be queried. The job is to give agents a structured, queryable path to the answer, and the instructions to use it.
Two halves — and we’re honest about both
① Discovery — partly out of our hands
“How does the agent find Rack Attack at all?” That’s pre-training + the agent’s own web search. No vendor fully controls it. What we can do is make the store legible once an agent arrives.
② Instruction + action — we deliver this
Once an agent is there: does it know how to get the right part, and can it act? This is the readiness layer Teifi builds and ships.
This is the “agentic readiness” story: stop promising magic discovery; own the part you control and do it better than anyone in the vertical.
Shopify’s agentic surfaces
/agents.md · /llms.txt · /llms-full.txt
The instruction layer. Shopify now auto-serves agents.md;
merchants customize it via templates/agents.md.liquid. It tells a visiting
agent what the store is and how to use it.
MCP server
The capability layer. A queryable tool surface the agent connects to — “find the part that fits this vehicle,” “add to cart.” Hosted, public, versioned.
agents.md teaches. MCP does. Together they turn a catalog from “scrapeable” into “operable” by an agent.
The fitment delivery, for parts merchants
Fitment MCP
One capability — vehicle → exact fitting product — exposed to ChatGPT, Claude & Copilot.
Authored agents.md
Encodes the YMM + disambiguation flow so agents ask the right question, not a long form.
Three demos, escalating
- On-site agent — “2020 Tacoma roof rack?” → it asks “naked roof or raised rails?” → returns the exact Thule / Yakima / Rhino kits.
- External agent — add the MCP URL as a Connector in ChatGPT / Claude; a third-party agent answers the same question using Rack Attack’s API.
- Checkout handoff — “add to cart” returns a real
checkout.rackattack.comlink via the Shopify headless API.
Same fitment tools power the storefront UI, the embedded agent, and the public MCP — one code path.
Where agents fail without help is where the moat is
Combinatorial complexity
SKU × vehicle × configuration. Exactly the space pre-training and scraping get wrong.
Fitment data is the moat
The merchant (and Teifi) own the authoritative fit graph. That’s the defensible asset in an agentic market.
RealTruck, DECKED, Rack Attack — every fitment-heavy catalog faces the same problem. Solving it once productizes across the vertical.
Agentic readiness for Shopify parts sellers
Audit
How legible & operable is the catalog to agents today? Structured data, agents.md, fitment coverage.
Build
Structured catalog · authored agents.md/llms.txt · hosted fitment MCP · on-site agent · checkout handoff.
Everything an agent buys is a product & variant
External agents shop a Shopify store through its Catalog API — and it speaks one language: products and variants. Fitment only becomes agent-usable once it maps onto those records.
Variants as fitment rows (DECKED)
Each fitment combo is a variant. Works with the Catalog API out of the box — but variant counts explode and the YMM logic stays implicit.
Structured YMM + fitment graph (Teifi)
Fitment lives in a graph that resolves to the right products/variants — cleaner catalog, explicit logic, agent-ready.
Teifi Parts
Shopify · synced by Teifi Bridge
Let’s make your catalog
agent-ready.
Unified commerce for complex brands. We deliver the instruction + action layer that lets AI agents find the right part and buy it.
teifi.com · Live demo: rackattack-fitment-demo.vercel.app