WebMCP: Why Nexbid Is Now Discoverable by Browser AI Agents
WebMCP makes websites executable for AI agents. Nexbid registers its 5 commerce tools directly in the browser — no screenshots, no DOM parsing.
Until recently, AI agents in the browser could only do one thing: look at web pages. Take screenshots, parse the DOM, guess what is clickable. That consumes 2,000+ tokens per interaction and achieves roughly 45% task accuracy. Now, Google and Microsoft have introduced a new browser API under the W3C: WebMCP (Web Model Context Protocol). It allows websites to expose structured, callable tools for AI agents. Nexbid is one of the first commerce systems to implement this protocol. What is WebMCP? WebMCP extends the browser with a new property: navigator.modelContext. Websites register tools in this context. Any AI agent in the browser — whether Perplexity Comet, OpenAI Atlas, or Chrome Auto Browse — can discover and directly execute these tools. The difference from screenshot-based interaction: Dimension Screenshot-based WebMCP Token consumption 2,000+ per action 20–100 per action Task accuracy ~45% ~98% Speed Slow (render + parse) Instant (JSON Schema) Reliability Fragile (layout-dependent) Robust (schema-defined) What Nexbid registers Nexbid exposes five commerce tools via WebMCP: searchOffers — Product search with category, budget, and intent filters getOfferDetails — Full product details including price, availability, and rating listCategories — All available product categories with counts initiateTransaction — Start a purchase (requires user confirmation) getOrderStatus — Check status of a pending transaction A browser agent visits nexbid.dev, queries navigator.modelContext, and immediately receives a structured list of all available actions. No crawling, no guessing, no DOM parsing. Dual-protocol: Backend MCP + WebMCP Nexbid now operates both protocols in parallel: Backend MCP (Anthropic protocol) — for CLI agents, API integrations, server-to-server communication. Available at mcp.nexbid.dev. WebMCP (browser API) — for browser agents operating in the user's context. Session cookies are inherited automatically. Both protocols serve the same commerce infrastructure. The difference is the transport layer: JSON-RPC over HTTPS vs. navigator.modelContext in the browser. Why this matters for advertisers and publishers Browser agents are already live. Perplexity Comet (since July 2025), OpenAI Atlas (since October 2025), and Chrome Auto Browse (since January 2026) can execute autonomous multi-step tasks in the browser. These agents search for products, compare prices, and trigger purchases. Without WebMCP, these agents see Nexbid products as pixels on a screen. With WebMCP, they see structured commerce tools with clear capabilities. That is the difference between looking at a page and using a page. Privacy by design WebMCP operates in a SecureContext (HTTPS-only), respects same-origin policy, and requires requestUserInteraction() for sensitive actions like transactions. The agent …
- Author
- Holger von Ellerts
- Published
- 2026-03-23
- Topics
- Agentic, Agent Discovery, Open Source