Marketfront AI
Marketfront AI is the commerce grounding layer for next-generation AI agents. It bridges the gap between conversational intent and physical reality, enabling your LLM to check live inventory, build validated carts, and execute real-world delivery orders with deterministic reliability.
Beyond Browsing: Actionable Commerce
We provide the "Do" capability for your AI:
- Deterministic Execution: Unlike scraping, our structured APIs ensure that when your agent adds a "Burger" to the cart, it's a real, purchasable item with a specific price and ID.
- Hallucination-Free Pricing: Live, grounded data means your agent never invents a discount or quotes an expired price.
- Contextual Grounding: Our search understands how food ordering works — filtering results not just by semantic relevance, but by real-time logistics (open hours, delivery radius, driver availability).
Two ways in
Marketfront AI ships two integration surfaces. Pick based on how tightly coupled you want to be to our runtime:
| Surface | Best for | Transport |
|---|---|---|
| Ordering MCP | Host a conversational agent in Claude Desktop / ChatGPT / VS Code Copilot / Cursor. Zero runtime on your side. | Model Context Protocol over Streamable HTTP |
| Marketfront REST APIs | You run your own orchestration layer — classic function-calling / tool-use against a typed REST surface. | HTTPS + Bearer JWT |
Both surfaces share the same backend (the Marketfront API) and the same session/auth model. Anything you can do through one, you can do through the other.
Ordering MCP — what you get
One connection, thirteen typed tools grouped by workflow stage:
- Fulfillment:
confirmFulfillment— shows the address on file, elicits pickup vs. delivery. Always the first call. - Discovery:
discoverStores,browseMenu,searchMenu,getItemOptions— find stores and explore their menus. - Cart:
addToCart,viewCart,updateCartItem,removeFromCart,clearCart— build and mutate the session cart. - Checkout:
reviewOrder,confirmPayment,placeOrder— compute totals, confirm the saved payment method, then submit.
Tools enforce stage ordering at the handler level: calling reviewOrder with an empty cart, or placeOrder without a prior confirmPayment, returns a structured error with agentGuidance for recovery.
Plus rich widgets (MCP Apps) — sandboxed iframes render in supporting clients at key steps (store discovery, menu browsing, cart review, order confirmation), so users see exactly what they're about to pay for.
Getting started
- Quickstart — connect an MCP client in ~3 minutes.
- Code Mode — why and when to let the model write JS instead of calling tools one-by-one.
- Tools reference — every tool, input schema, and output shape.
Need access? Contact our partnerships team for API keys.