WebMCP

AI assistants have started browsing websites on people’s behalf. Someone asks ChatGPT to find them a personal loan, and its built-in browser opens your site, reads the page and tries to fill in your form the way a person would. It’s slow. One unclear field or an unexpected pop-up is often enough for the agent to give up, and the visitor never reaches you. WebMCP is a new web standard that lets your website tell the agent what it can do there, like “get a loan quote” or “book a call with an advisor”. The agent then asks for that action directly instead of clicking around. With Waniwani, those actions are the ones your agent already offers in ChatGPT, Claude and the chat on your site, so an agent on your website asks the visitor the same questions in the same order, and every lead lands in the same analytics.

​

If the Waniwani chat is on your site, WebMCP is already on

WebMCP comes built into the Waniwani chat. Every page that loads the chat with the snippet from your dashboard already offers your agent’s actions to AI agents. There’s nothing to install or switch on. Ordinary visitors see no difference. The chat looks and behaves as it did, and the browsers most people use today don’t support WebMCP yet, so for them it stays dormant and loads nothing extra.

​

Add it to a site without the chat

You need a live Waniwani agent with a website channel. The snippet you paste is the same one that installs the chat, and you choose whether the chat shows. If your developers built the chat into your site with Waniwani’s React component rather than the snippet, WebMCP isn’t included. The developer guide shows them how to add it.

​

What agents can do on your site

An agent browsing your site gets the actions your Waniwani agent offers in ChatGPT and Claude, and only those. If your agent gives loan quotes and books advisor calls, those are the two things an agent can do on your website. Everything else on your site it reads like any other page. Some steps show something visual, like a quote card or a calendar for picking a slot. On your website these open in a window over the page, so the visitor sees it and makes their choice, and then the agent carries on from there. When your agent gains a new action, agents on your website get it within a couple of minutes. Your website doesn’t change.

​

Making a website form available to agents

WebMCP doesn’t pick up the forms already on your pages. To let agents complete what a form does (a loan application, a callback request, a quote), that form becomes a guided flow inside your Waniwani agent, which asks the visitor each question in conversation and sends the answers wherever your form sends them today. Once the flow exists, it works in every channel at the same time. Agents browsing your site, ChatGPT, Claude and your website chat all get it, and nobody touches your website. To build it, the Waniwani team needs four things from you:

• A link to the form, or its list of fields with the required ones marked.
• Where submissions go today (a CRM, an inbox, an internal system) and access to send them there.
• The rules the form enforces, such as a minimum loan amount or an age limit.
• What the visitor should see after submitting: a confirmation, a next step or a link.

If you build your agent yourself with the Waniwani SDK, the developer guide has a worked example.

​

Control where it runs

Set Allowed origin on your website channel’s Settings tab, for example https://www.example.com, and Waniwani only answers pages on that address. example.com and www.example.com count as two different addresses, so enter the one your visitors land on. To leave the chat on a page and keep agents from using your actions there, your web team adds data-webmcp="false" to the snippet on that one page. Other pages keep working as before.

​

Measure what agents bring in

Everything an agent does through WebMCP is recorded in your Waniwani analytics under your website channel, alongside your chat, ChatGPT and Claude numbers. When a visitor starts with an agent and later continues in your chat, both count toward the same person.

​

Which AI agents use it

WebMCP is new. ChatGPT’s desktop app uses it when its browser visits a site for someone, and Google is testing it in Chrome. Agents that don’t support it keep reading and clicking through your pages the usual way. Because it ships inside the chat snippet, each agent that adds support can use your actions from day one, with no change on your side.

​

For developers

How it works under the hood, how to test a page before agents reach it, troubleshooting and installing without the snippet are all in the WebMCP developer guide.