Making a fitment-heavy catalog usable by AI agents
Buyers are starting to shop through AI assistants. For parts & automotive, the first question is always “will this fit my vehicle?” — and that's exactly what agents get wrong. Here's the problem, and the three surfaces that solve it.
Two halves — and we're honest about both
① Discovery — partly out of our hands
How does an agent findRack Attack at all? That's the model's pre-training plus its own web search at query time. No vendor fully controls it. What we can do is make the store legible the moment an agent arrives.
② Instruction + action — we deliver this
Once an agent is here: does it know how to get the right part, and can it act? That's the agentic-readiness layer — and it's entirely buildable.
How an agent “sees” a product
The first two are guesses. Fitment can't be guessed — it must be queried. Our job is to give agents a structured, queryable path to the answer, plus the instructions to use it.
The disambiguation problem
A 2020 Toyota Tacomadoesn't map to one roof rack. It maps to four — depending on the roof type:
A good agent must detect the missing variable and ask— “naked roof or raised rails?” — instead of guessing. Try it:
The three surfaces
1 · Instruction surface — agents.md teaches
Tells a visiting agent what the store is and how to query fitment. Shopify now auto-serves agents.md; merchants customize it via templates/agents.md.liquid. This is exactly that file, hand-authored for fitment.
2 · MCP capability — the server does
A public Model Context Protocol server exposes the fitment tools. Any external agent — ChatGPT, Claude, Copilot — can connect and answer “what fits my vehicle?”
3 · On-site agent — proof in context
An embedded Claude assistant (bottom-right) runs the same tools to resolve fitment and recommend real products, then hands off to the real checkout.rackattack.com checkout.
Powered by Teifi
Two purpose-built Teifi products turn a complex fitment catalog into a catalog an AI agent can query in real time — at the scale and freshness merchants actually need.
Teifi Parts
YMM elastic search across millions of SKUs
Instant, typo-tolerant fitment search at full-catalog scale. Every vehicle-finder dropdown, every agent tool call, and every MCP query on this demo resolves through Teifi Parts — returning the exact rack systems that fit a specific Year / Make / Model / body / roof configuration in milliseconds.
- ·Elastic search handles brand misspellings, abbreviations, and partial queries
- ·Fitment graph covers YMM + body type + roof / hitch / bed configuration
- ·Structured JSON responses are designed for agent and MCP consumption
Teifi Bridge
Two-way sync from PIM / ERP / data lakes
Keeps the product catalog and the fitment graph continuously in sync from the merchant's source systems. Bridge ingests from any PIM, ERP, or data lake — normalises it — and pushes structured fitment data into Teifi Parts. It also hosts the whole experience: this storefront, the MCP server, and the agent.
- ·Connectors for Inriver, Jitterbit, Shopify, and custom REST / GraphQL sources
- ·Bidirectional: enriched fitment data flows back to the merchant's PIM
- ·Hosts the MCP server, the on-site agent, and the storefront front-end
On the roadmap
Connect the merchant's customer-service platform and internal knowledge base so the assistant answers from their own fit advice, install guides, and support history — the same agent, now backed by the team's expertise. Available to wire in; not enabled in this demo.
Connect an external agent (live)
- ChatGPT / Claude → add a custom Connector (MCP) with the URL above.
- Ask: “Using Rack Attack, what roof rack fits a 2020 Toyota Tacoma with a naked roof?”
- The external agent calls this server's tools and returns the exact fitting kit.