Your software doesn't talk to AI agents yet.
More and more people get their work done through AI assistants like Claude, ChatGPT or Copilot. If an assistant is supposed to complete a task in your product, today it has to click its way through an interface built for humans, screenshot by screenshot. That is slow, error-prone and breaks with the next design change.
Agents don't need a nicer interface, they need tools: clearly described actions like “create quote” or “reschedule appointment”, with fixed parameters and unambiguous results. That is exactly what the Model Context Protocol (MCP) is for, an open standard through which assistants like Claude or ChatGPT can use your product's features directly.
WebMCP brings the same idea to the browser. It is an emerging web standard that browser vendors are developing together in a W3C group and that Chrome is already testing. Your web application registers its tools right in the browser: an agent uses the user's existing session, and the person sees in the familiar interface what the agent is doing.
Does this sound familiar?
Customers ask for an AI integration, and nobody knows exactly what that means.
A chatbot widget was added, but hardly anyone uses it.
There is an API, but AI agents don't understand when and how to use it.
What waiting costs you every month
As long as your product offers no tools for agents, your customers get their tasks done where it's easier, often with a competitor. Agents clicking through interfaces also cause errors and support requests that reflect on your product. And whoever doesn't start now will later rebuild under time pressure what others have long since tried and tested.
How I approach it
Shape the tools
Together we work out which tasks agents should take over in your product and which they shouldn't. The result is a small set of clearly cut tools with clear permissions. That is interface design, just for a new kind of user: the agent.
Build MCP servers and WebMCP
I implement the tools as an MCP server for assistants like Claude and ChatGPT and, where it fits, as WebMCP tools right in your web application. That includes proper sign-in via OAuth, understandable descriptions and confirmation by a human before critical actions.
Test with real agents
Automated test runs with real AI models show whether agents choose and use the tools correctly. Logs reveal what agents actually do, and safeguards prevent manipulated content from tricking an agent into unwanted actions.
Frequently asked questions
What is the difference between MCP and WebMCP?
An MCP server runs on your server and provides tools for AI assistants, independent of the browser. WebMCP runs in your web application in the browser: the agent works with the signed-in user's session, and the user watches along. Many products benefit from both.
Is WebMCP ready for production?
WebMCP is still a draft. Chrome is already testing it, and other browsers take part in its development. That's why I build it as an addition that doesn't affect today's users. MCP, on the other hand, is established and supported by the major assistants.
Do we need a new API for this?
Usually not. The tools build on your existing API or backend. What matters is how they are cut: a few understandable actions instead of every single endpoint.
How safe is it to give agents access?
As safe as the tools are built. Agents only get the permissions of the respective user, critical actions require confirmation, and every action is logged. Special attention goes to prompt injection, meaning content designed to lure an agent into unwanted actions.
Does this replace our chatbot?
Often, yes. Instead of a chatbot of your own that users first have to find and figure out, your product works with the assistants your customers already use.
Can't an AI agent simply operate our application by itself?
It can try, using screenshots and simulated clicks. That is slow, expensive and fails on details like a moved button. The agent guesses what your interface means, with clearly described tools it knows. Designing those tools is design and development work at the same time, and that is exactly where my focus lies.
Let's talk about your situation
In a free initial call we take a look at your problem together. You get an honest assessment of whether and how I can help.