One of the most interesting trends in Search over the last few years has been the rise of ChatGPT, Claude, Gemini and more. TV voice searches for products are now commonplace, embedded offers across sales platforms like Amazon, eBay, Expedia, Autotrader and more are often working better in terms of conversion per enquiry, than traditional website advertising or spamming local news media sites.
For insurers the future is undoubtedly data driven, maybe not specific consumer query driven. So much data on renewal dates, holidays booked, MoT or DVLA tax expiry etc is out there and can be used by AI to prompt policyholders, regardless of the media they are using at the time. It’s all about hitting the sweet spot on home or car renewal, sending a nudge out via a shopping app, Instagram, FB or TikTok can be a call to action. Welcome to the future.
Here’s some news from Waniwani;
Since February, AI assistants have been able to quote a real insurance policy. 128 financial services apps now run on ChatGPT, from Aviva and Liberty Mutual to Allianz and BBVA, but they only reach customers who proactively know which AI app to ask for. Claude from Anthropic solved the problem by suggesting the right app in conversation. But for the mass market using ChatGPT and Gemini, users need to name the app or install it in settings, which prevents mass usage.
WebMCP changes that. The standard Google and Microsoft introduced and OpenAI adopted in August removes the app friction. An agent browsing a website can ask for a price and get one. Tuio, Pretto and Lassie are the first financial services companies to switch it on, through Waniwani infrastructure.
In June, for the first time in the history of the internet, more of the web’s traffic came from machines than from people. Cloudflare measured it: 57.5% of requests for web pages were automated, against 42.5% from humans.
Almost none of those websites were built to be read by a machine. Financial services websites least of all, because what a customer wants from an insurer or a broker is an offer with a price, and that offer and price have to be computed after an (often long) form.
The institutions have already adopted. The customers never found them.
128 apps on ChatGPT can quote, apply for or start buying a financial product, according to Waniwani’s review of the directory. Seventy-four are insurance, and 71 of those return a real price.
The list runs through the industry’s largest names. Aviva launched a home insurance app in April and extended it to life cover in June. Liberty Mutual became the first major US carrier to quote auto insurance inside ChatGPT. Allianz quotes motor cover. In the US, Ethos writes term life estimates, Insurify compares auto quotes, and Intuit’s TurboTax returns tax estimates. In the UK, Compare the Market and MoneySuperMarket both run car insurance apps. In banking, BBVA runs a conversational app inside ChatGPT for customers in Italy and Germany.
Adoption among institutions ran far ahead of adoption among customers, because an app reaches a customer only when that customer already knows it exists.
What has been missing is a way for those agents to ask
Scraping a page returns a marketing claim, never a quote. Until now the only route to a real quote was an app inside an assistant, and that reaches a customer only if the customer already knows the company by name and asks for it. Claude closed that gap in May 2026 by suggesting relevant apps as a conversation unfolds, from a directory of more than 200. ChatGPT and Gemini, which between them carry the overwhelming majority of consumer AI use, still leave discovery to the user: ChatGPT paused its own contextual app suggestions after backlash over recommendations that read as advertising, and Gemini’s Connected Apps must be authorised one by one.
WebMCP is the ignition
A website publishes a machine-readable list of things it can do, and any agent that lands on the page can call them directly. The agent needs no app, no app store and no prior knowledge of the company. It reaches the tools by visiting the site it is already visiting when performing websearch. But with WebMCP it now has a door dedicated to agent visitors only, which can return a quote once the inputs are provided. WebMCP solves the discovery and suggestion problem of AI apps. The agent can speak and interact directly with the website.
Today, Tuio (leading digital home and life insurance in Spain), Pretto (mortgage brokerage in France) and Lassie (leading pet insurance in Sweden) become the first financial services providers to expose their quoting tools to AI agents using WebMCP. An AI agent arriving at any of the three websites can now request and receive a personalised quote, priced by the company’s own engine, without a human ever filling in a form.
All three run on Waniwani infrastructure. The company announced the rollout of WebMCP to all its customers at no additional cost, and has added it to its open-source framework for MCP builds.
What WebMCP is, in plain terms
Most people in insurance have not heard of WebMCP. It is worth thirty seconds.
When an AI agent uses a website today, it does what a confused human would do: take a screenshot, guess which box is the postcode field, click, and hope. It is slow, expensive and wrong often enough to matter.
WebMCP replaces the guessing with an invitation. The website publishes a short, machine-readable list of things it can do (get_home_insurance_quote, here are the five inputs I need, here is what I will return) and the agent calls it directly.
The useful comparison is structured data and Google search. Twenty years ago, search engines had to infer what a page was about. Then websites started declaring it: this is a product, this is a price, this is a review. Search got better for everyone. WebMCP does the same thing for actions.
It was co-authored by Google and Microsoft, and is being developed in the W3C’s Web Machine Learning Community Group. Google opened a public trial in Chrome in May. Since August it has been switched on by default across Shopify storefronts and Cloudflare-fronted sites, covering millions of pages. On 25 August, OpenAI enabled WebMCP support in the ChatGPT desktop app’s built-in browser, so that ChatGPT Work and Codex can discover and use the tools a website exposes.
Retail and travel moved first. Expedia, Instacart and Target are among those experimenting. Financial services was absent from that list until today.
Discovery stops being the bottleneck. This is the part the industry has underestimated. Since February, vendors have been able to deliver personalised quotes inside a user’s conversation by deploying an MCP server as a ChatGPT AI app (now called ChatGPT Plugin), a shift the market priced immediately by removing some $79bn of insurance broker and insurer stock value in a single day according to Goldman Sachs. It works, with one catch: it only reaches customers who already know the company exists inside their assistant and ask for it by name. ChatGPT’s own mechanism for solving that, where the assistant suggests the right app at the right moment, was paused after user backlash and has been unreliable since. WebMCP takes the app store out of the path entirely.
It is auditable. Because the agent calls a declared tool instead of impersonating a mouse, the company chooses which actions to expose, and can see exactly what was called, by which agent, and what came back. That matters in a regulated industry.
For the vendor, the work is marginal. The same MCP built for a ChatGPT app is exposed on the website through a script. Nothing changes for human visitors.
Seven months ago, MCP was the first step towards agentic distribution in financial services. WebMCP is the second, and its impact will be larger, because every website already receiving agent traffic can now quote with it. For an AI agent, a website has stopped being a static brochure and become a sales channel.
“We always saw MCP as the first step in a much bigger shift in how insurance gets distributed, and WebMCP confirms it. We built the first insurance app on ChatGPT with Tuio, so it mattered to us to be the ones putting the first WebMCP into production as well. It is live today for Tuio, Lassie and Pretto, it will roll out to the rest of our customers, and it goes into our open-source framework, because this needs to become how the industry works rather than a feature we own. What it does is turn a website that was optimised for humans into one that is optimised for agents.”
Robin Diligent, CEO and co-founder, Waniwani
Be the first to comment