WebMCP Support in Salespeak: How AI Agents Can Now Ask Your Website Anything

Google just released WebMCP in early preview. It's a new standard that lets websites expose structured tools directly to AI agents — no more brittle DOM scraping or clumsy browser automation.

And Salespeak already supports it.

We moved fast on this because WebMCP is exactly the future we've been building toward. For months, we've been talking about agent-first web design and building infrastructure for a web where AI agents are first-class visitors. WebMCP is the standardization layer that makes that vision practical.

What Is WebMCP?

WebMCP is Google's proposal for an "agent-ready" web. The core idea is simple: instead of AI agents reverse-engineering your website's UI — clicking buttons, filling forms, trying to make sense of navigation menus designed for humans — your site exposes structured tools that agents interact with natively.

Think about the difference between screen-scraping a website and calling an API. One is fragile, slow, and breaks every time you redesign a page. The other is fast, reliable, and precise. WebMCP brings that same leap to how AI agents interact with websites.

The standard provides two complementary approaches:

• A Declarative API that enables standard actions defined directly in HTML. Your existing forms, search bars, and product selectors become agent-accessible tools without writing new code. The agent reads your HTML annotations and knows exactly what tools are available.
• An Imperative API for complex, dynamic interactions requiring JavaScript execution. This is where multi-step workflows live — the kind of back-and-forth conversations that actually matter in B2B, where a buyer agent needs to ask follow-up questions based on previous answers.

Together, these give websites a structured way to communicate capabilities to any AI agent that visits. No guesswork. No brittle automation scripts that break when you move a button three pixels to the left.

Under the Hood: How WebMCP Tools Work

Here's where it gets interesting technically. WebMCP tools follow a familiar pattern if you've worked with function calling in LLMs or the Model Context Protocol. Each tool exposed by your website has a name, a description, and a defined parameter schema. When an AI agent lands on your page, it discovers these tools through the WebMCP manifest and can invoke them programmatically.

A simple declarative tool might look like a search endpoint — the agent passes a query string and gets structured results back. But the imperative API opens up something more powerful: stateful, multi-turn interactions where the tool can ask clarifying questions, maintain context across exchanges, and drive a real conversation.

This is the technical foundation that makes agent-to-agent commerce possible. It's not just "expose an API." It's "give AI agents a way to have intelligent, structured conversations with your website."

How Salespeak Implements WebMCP: One Tool, Infinite Conversations

Here's where our approach gets opinionated — and we think, elegant.

Most companies thinking about WebMCP will probably start listing out dozens of tools: one for pricing lookups, one for feature comparisons, one for checking integrations, one for booking demos. That's the instinct. Map every page and function to its own tool.

We went the other direction. Salespeak exposes a single WebMCP tool called ask_question.

That's it. One tool. It takes a question string as its parameter. The tool definition includes a name (ask_question), a description that tells the visiting agent what it can ask about, and a simple parameter schema with a single required field: question, a string. That's the entire interface.

Why a single tool? Because the real value isn't in exposing structured endpoints for every possible query. It's in having an intelligent agent on the other side that understands context, handles ambiguity, and drives conversations that actually qualify buyers.

When a buyer's AI agent calls ask_question with "What integrations do you support for enterprise CRM platforms?", it doesn't get a static JSON dump of integration names. It gets a response from your Salespeak agent — trained on your actual product knowledge — that answers the question and might ask back: "Which CRM is your team currently using? That'll help me give you the most relevant integration details."

The buyer's agent responds. Your agent qualifies. The conversation continues through structured protocol, each exchange building on the last. No DOM parsing. No guessing which button to click. Just two AI agents having a productive conversation about whether your product is the right fit.

Why This Design Decision Matters

A proliferation of narrow tools creates a discovery problem. An AI agent landing on your site sees 40 tools and has to figure out which ones to call and in what order. That's cognitive overhead — for an AI agent, yes, but it's real overhead nonetheless. Tool selection errors lead to wasted calls, incomplete information, and poor recommendations back to the human buyer.

A single ask_question tool eliminates that problem entirely. The agent knows exactly what to do: ask a question. The intelligence lives in your Salespeak agent, not in the tool schema. Your agent handles routing, context management, and qualification logic internally — the visiting agent just needs to have a conversation.

It's the same reason humans prefer talking to a knowledgeable salesperson over navigating a 47-tab product wiki. The interface is simple. The intelligence is behind it.

Why This Matters for B2B

Here's what's already happening: a buyer asks their AI agent to research solutions for their team. The agent visits vendor websites, evaluates options, and reports back with recommendations.

Today, that agent is parsing your DOM. It's reading your marketing copy, trying to extract structured information from content designed for human eyes. Sometimes it gets things right. Sometimes it misreads your pricing page, confuses your product tiers, or misses your strongest differentiator because it was buried in a carousel.

With WebMCP and Salespeak's ask_question tool, that same agent connects to your website and simply asks what it needs to know. It gets accurate, contextual answers from an agent that knows your product inside and out. And your agent qualifies the opportunity in the process.

The difference between those two experiences is the difference between getting recommended and getting overlooked.

Agent-to-Agent Commerce Is Real

When a buyer's AI agent lands on your site, it shouldn't encounter a static form and a chatbot that deflects to "check our FAQ." It should encounter an intelligent counterpart — your Salespeak agent — that can hold its own in a real conversation.

That's agent-to-agent commerce. The buyer's agent calls ask_question. Your agent answers with depth and accuracy. The buyer's agent follows up. Your agent qualifies the fit and captures the lead. All through structured protocol, all generating structured data you can analyze.

This isn't theoretical anymore. WebMCP is the infrastructure. Salespeak is the intelligence layer. The companies that wire these together first will have a meaningful head start.

What You See on Your End

Every ask_question call through WebMCP generates structured interaction data that flows into your Salespeak dashboard. You see exactly what AI agents are asking about, what follow-up questions they pose, and how your agent qualifies them.

This is visibility into a buying process that most B2B companies are completely blind to. You can see which topics come up most in agent-to-agent conversations, which qualification questions drive the most engagement, and where agents drop off. It's the kind of intent data that used to require a human SDR on a phone call — now captured automatically, at scale, 24/7.

You can query all of this through the Salespeak MCP Server too — ask your own AI assistant "What are buyer agents asking about most this week?" and get an answer in natural language.

What You Should Do Right Now

WebMCP is in early preview. The standard will evolve. But the direction is clear: AI agents are going to interact with websites through structured tools, not browser automation. Here's how to get ahead of it:

1. Make Your Website Agent-Ready

If your website still treats every visitor as a human, you're building for yesterday. AI agents are already visiting your site — they're already trying to extract information. WebMCP gives them a better way to do it, but only if you expose the tools.

Start by understanding how agents currently see your site. What can they extract? What are they missing? If ChatGPT can't accurately describe your product after visiting your homepage, you have a problem that WebMCP alone won't solve — but it's a problem Salespeak is built to fix.

2. Deploy a Conversational Agent

WebMCP makes conversational interfaces the primary way agents interact with your site. A static FAQ page doesn't cut it. You need an intelligent agent that can handle open-ended questions, qualify visitors, and provide accurate product information — whether the visitor is human or AI.

Salespeak does both. Our conversational AI is trained on your actual product content, handles human visitors through the widget and AI agents through WebMCP's ask_question tool, and qualifies every interaction regardless of who — or what — is on the other end.

3. Think Agent-to-Agent

The web is moving from human-to-website to agent-to-agent. Your GTM strategy needs to account for AI agents as a primary audience. That means rethinking inbound — not just optimizing for the humans who land on your site, but for the AI agents researching on their behalf.

This is a mindset shift. Your website isn't just a brochure anymore. It's a participant in conversations you may never see directly. Make sure it's saying the right things.

The Agentic Web Is Here

WebMCP is one of several signals that the web is fundamentally changing. Between Google's WebMCP, Microsoft's NLWeb, and Anthropic's MCP, the infrastructure for agent-to-agent communication is being built in real time.

Companies that adapt now get compounding advantage — every agent interaction trains your system, every structured conversation generates data, every successful qualification builds your position in the agent-mediated purchasing process. The companies that wait will eventually catch up on the technology. They won't catch up on the data.

Salespeak makes your website agent-ready today — with WebMCP support and a single, powerful ask_question tool that turns every AI agent visit into a qualified conversation.

Book a demo and we'll show you what it looks like when AI agents can actually talk to your website.