Live today, not a roadmap.
Model Context Protocol is how AI agents check your inventory, hours, and service availability without scraping your site. PromptGraph runs a per rooftop MCP server for every client store right now, at mcp.promptgraph.ai.
The problem MCP solves
When a shopper asks ChatGPT "does anyone near me have a white F-150 Lariat under $60k," the model has three ways to answer. It can guess from training data, which is months old. It can search the web and read whatever it can parse, which on most dealer sites means fighting a JavaScript inventory grid. Or it can call a tool that returns the answer as structured data.
MCP is the open standard for that third option. Anthropic published it in late 2024. A dealer exposes a small server with a handful of tools. An AI agent connects, asks, and gets JSON back. No scraping, no guessing, no stale price.
What a PromptGraph dealer server exposes
Inventory search
By year, make, model, trim, price range, body style, and drivetrain. Returns live units with stock number, VIN, price, and a link to the VDP.
Vehicle detail
One VIN in, full record out: photos, options, and window sticker data where the feed provides it.
Store profile
Name, address, phone, hours by department, brands sold, and the Google Business Profile link so the agent can match your store to the entity it already knows.
Service
Service hours, the services you offer, and a scheduling link the agent can hand to the shopper.
Why this matters more than schema
Schema markup describes your store. MCP lets an agent act on it. In our September 2026 scan of franchise dealer websites, almost every store had AutoDealer schema and almost none had an agent endpoint. Schema is table stakes. The agent layer is the whitespace, which is why it carries 55 of the 100 points in our readiness score.
How it goes live
We connect your inventory feed
The server runs on PromptGraph infrastructure and reads the feed you already have. No IT project on your side.
We register the server against your domain
The manifest is published at the well known path and your llms.txt and discovery file point to it, so agents can find the store from your own site.
You watch an agent use it
On the demo we connect an agent to your rooftop and ask it real shopper questions. Then we track what ChatGPT, Gemini, and Grok say about your store in your dashboard.
Straight answers
Do I need to change my website platform?
No. The server runs separately from your website and reads your inventory feed. Dealer.com, DealerOn, Dealer Inspire, and OEM mandated sites all work.
Is MCP the same as llms.txt?
No. llms.txt is a text file that tells a language model what your site is about. MCP is a live server an agent can query for current data. You want both. They are two different checks in the readiness score.
What does it cost?
Nothing beyond the PromptGraph subscription, $399 per rooftop per month. The MCP server is included.
What does it not do yet?
No consumer assistant auto discovers every dealer MCP server on the internet today. Adoption is moving engine by engine. What is true today is that agents built on Claude, on OpenAI's tooling, and on a growing list of shopping assistants can connect to a published MCP server. The stores that are live get the traffic when each engine flips the switch.
Can I see a live dealer server?
Yes. Book a demo and we will connect an agent to a client rooftop while you watch.
See an agent query a real dealership.
Free AI visibility scan first, no card. Then a live MCP demo on a client rooftop.
Book a Demo See Pricing