7 min read 1,390 words

More and more people research with an AI assistant in their browser. So that this assistant does not have to guess on our site, freseniusinstruments.com has offered its own tools based on the new draft web standard WebMCP since 2 November 2026. An agent can query product data, measuring ranges, distribution partners and open positions directly instead of reading the page line by line. Here I explain what WebMCP is, how it works and where its limits are.

In short: what is WebMCP?

WebMCP (Web Model Context Protocol) is a draft by the W3C Web Machine Learning Community Group, with Google and Microsoft among those working on it. Through WebMCP a website registers tools in the browser: each has a name, a description, defined inputs and a function that returns the answer. An AI agent in the same browser sees these tools and calls them. It receives structured data instead of having to interpret text, tables and menus on its own. WebMCP is not an official standard yet; it is a draft that is still changing.

Why we use WebMCP

An agent that only reads a website works like someone copying a data sheet through frosted glass. It recognises a lot, but not everything. A measuring range table with three units, a footnote about the hazardous zone, or whether there is a distribution partner in a given country: this is exactly where mistakes happen.

In gas analysis, these details matter. Whether an analyzer measures H₂S in the ppm range or in the percent range decides which one fits. With WebMCP the agent gets the values from the same source our product pages are built from. What you read on the page and what your assistant tells you are the same.

How WebMCP works

1. The page describes its tools

When it loads, our page checks whether the browser supports WebMCP. Only then does it register its tools; otherwise it loads nothing extra. Each tool explains in one sentence what it does and defines the inputs it expects, such as a search term or a country code.

2. The agent picks a tool and calls it

When you ask your assistant a question, it reads the list of tools and decides which one fits. For “Which analyzers measure oxygen?” it calls the product search with the term “oxygen”, for example.

3. The answer comes from the same data as the page

The tool queries the same data as our pages and returns the result in a structured form: name, short description, technical data and the link to the product page. The agent builds its answer from that and can send you straight to the right page.

For the technically minded, this is the core of how a tool is registered (shortened):

const mc = document.modelContext;
if (mc && typeof mc.registerTool === 'function') {
  mc.registerTool({
    name: 'search_products',
    description: 'Search the Fresenius Instruments product range ' +
      '(gas analyzers and sample gas coolers).',
    inputSchema: {
      type: 'object',
      properties: {
        query: { type: 'string', description: 'Search words, e.g. "CH4"' }
      }
    },
    annotations: { readOnlyHint: true },
    async execute({ query }) {
      const url = '/wp-json/tory-seo/v1/agent/products?q=' +
        encodeURIComponent(query || '');
      const data = await (await fetch(url)).json();
      return { content: [{ type: 'text', text: JSON.stringify(data) }] };
    }
  });
}

The readOnlyHint annotation tells the agent that the tool only reads and changes nothing.

The seven tools on freseniusinstruments.com

  • Search products (search_products): gas analyzers and sample gas coolers by keyword, gas component or application.
  • Product details (get_product_details): measuring principle, gas components and ranges, accuracy, interfaces, the FAQ of the product page and binding notes. For the BioBasic, for example, it states that the analyzer itself is not ATEX-approved and is installed outside the hazardous zone.
  • Substances (search_substances): around 120 substances with measuring principle and measuring ranges per unit, as in our substance database.
  • Find distribution partners (find_distribution_partners): partners by country. If there is none, the tool points you to us directly.
  • Prepare a partner inquiry (prepare_partner_inquiry): creates a link to the inquiry form with your request filled in. Nothing is sent until you click “Send” yourself.
  • Open positions (list_open_jobs): our current vacancies with location, start date and a link to the careers page.
  • Contact options (get_contact_options): which page is right for which request, from general questions to quotes, service and booking a call.

In addition, the partner inquiry form is described so that an agent can pre-fill its fields. Consent and submitting stay with you.

A practical example

Say you run a biogas plant in Italy and ask your assistant: “Which Fresenius Instruments analyzer measures methane and hydrogen sulphide in biogas, and who looks after customers in Italy?”

  1. The agent calls the product search with “biogas” and finds the BioBasic, among others.
  2. Product details give it the measuring ranges and the notes, including the one on the hazardous zone.
  3. The partner search for Italy returns the responsible distribution partner with a link to their profile.
  4. If you wish, it prepares the inquiry to this partner. You check the text and send it yourself.

Every step relies on our own data. The agent neither copies tables nor hunts for country names in a map menu.

In the partner portal too: tools for our partners

Our partner portal Insider, where partners manage leads, quotes and their status, offers WebMCP tools as well. What an agent may retrieve there depends on the role of the person who is signed in:

  • Everyone: the public partner search.
  • Applicants: in addition, their own onboarding status with the next step.
  • Partners: in addition, partner tier, progress and their own leads. The tool deliberately leaves out customers’ names and contact details.

An agent never has more rights there than the person who is signed in. You can read more about the programme in the partner programme announcement.

What an agent cannot do on our site

WebMCP also allows tools that trigger actions, such as placing an order. We do without them. All our tools only read:

  • No tool sends, orders or changes anything.
  • An agent may at most pre-fill forms. Submitting, consent and the anti-abuse check stay with the person.
  • The prepared inquiry link contains no names, e-mail addresses or phone numbers.
  • The website only returns data that is already public on our pages.

Where WebMCP works today

WebMCP is new, and support is only starting to grow. As of November 2026:

  • ChatGPT desktop app: its built-in browser finds a page’s tools and calls them. OpenAI calls this “Site tools”.
  • Google Chrome: WebMCP runs as an origin trial from Chrome 149. Developers can enable it via chrome://flags/#enable-webmcp-testing.
  • Other browsers and assistants: Safari and Firefox do not support WebMCP so far. Many assistants still read pages the conventional way.

That is why WebMCP is an addition for us. Every page works without WebMCP exactly as before, and an agent without WebMCP still finds all content in the text of the page.

Frequently asked questions about WebMCP

Will I notice anything as a visitor?

No. Without a browser that supports WebMCP, nothing extra is loaded. The pages look and load the same.

Can an AI agent order something or send an inquiry on your site?

No. Our tools only read. An agent can prepare an inquiry, but you have to send it yourself.

What is the difference between WebMCP and MCP?

The Model Context Protocol (MCP) connects AI applications to tools and data through a separate server. WebMCP brings this idea into the browser: the website itself provides the tools, and they run in the session you currently have open.

Is WebMCP an official standard?

Not yet. It is a draft by a W3C Community Group. Details may change, and we will adapt our tools when they do.

Conclusion

With WebMCP we give AI assistants direct, verified access to our product data, measuring ranges and contacts. The answers you get through your assistant come from our own data rather than from an interpretation of the page. The decision stays with you: only you can send, inquire or order.

Looking for the right gas analyzer? Browse our products, find a distribution partner near you or talk to our team directly, with or without an AI assistant.

Sources