From answering to acting: agents, (MCP) connections and widgets in practice
From showing information to getting things done: how agents, MCP connections and widgets let a website act.
Joris Meijer Co-founder, AI Lead
4 min read
A website that answers with AI can more and more often get something done for the visitor, too. A visitor says, for example: "I want to apply for a permit." Or: "Book me an appointment for next Tuesday." Or simply: "I want to place this order." The visitor then expects the website to start the process right away. That calls for a different way of thinking, and a different technical make-up of websites.
When language gets access to systems
As long as AI only generates text, the website stays informative. To get from conversation to result, language has to be able to work safely with the systems behind it: stock data, scheduling software, CRMs, internal processes.
That needs a standardized connecting layer. The Model Context Protocol (MCP) lets an AI retrieve data and prepare or carry out actions in a controlled way, within set business rules. The system can pull up an earlier order, check current availability and assemble a new basket, for instance, without the user having to work through several screens.
Websites have always been able to support processes; the difference is in the access to them. AI makes it possible to drive systems straight from the conversation, without the detour of traditional navigation and fixed flows.
More than reacting: how agents think
Besides access to systems there is another important change: the way questions are processed. It started with reacting to a single prompt. Now agents can work through several reasoning steps before they answer. They retrieve extra information, apply rules, compare options and decide what the logical next step is.
You can also run different agents at the same time. A service agent gets different instructions and has different priorities than a sales agent. An internal colleague needs a different approach than an external customer. Digital interaction becomes less generic and more tuned to the user's role.
Read more about agents in vragen.ai
When conversation turns visual
Even so, text on its own only goes so far when choices or transactions are involved. Which brings a third development: interactive blocks that appear in the conversation. Widgets make it possible to show products, calendars, status overviews or comparisons right there in the exchange.
The answer is no longer "Click here to continue." The answer is the next step. The user sees available time slots, adjusts quantities or confirms a choice, without leaving the conversation.
Read more about widgets in vragen.ai
Why the combination makes the difference
On their own, agents, system connections and widgets are interesting innovations. Together they change what a website is. Through the conversation, a visitor then books an appointment or places an order.
Here is how the three play together:
| Component | Role in the interaction |
|---|---|
| Agent | Understands intent, reasons in several steps and decides on the right next step |
| MCP | Connects the conversation to the systems behind it and makes controlled actions possible |
| Widget | Makes information and actions visual, interactive and immediately actionable inside the conversation |
When these three layers work together, a visitor asks for an appointment and picks a time slot right in the conversation. The user then no longer has to navigate menus someone designed in advance.
What this asks of your website
You could of course always go through a process on a website. Permits could be applied for, appointments booked, orders placed.
AI changes the access to these processes. The visitor says what they want and doesn't have to follow a fixed menu or pre-designed flow. The user states a goal, and the system works out the fastest route to carrying it out.
That calls for a different way of thinking. Next to the question of which page you show comes the question of which intents you want to support directly and which systems you need for that.
AI works as the connecting layer here. What used to be spread across pages and forms can now be brought together in one conversation.
Getting started with vragen.ai
At vragen.ai we make this concrete. Organizations can configure agents with specific roles and reasoning steps, set up MCP connections to their own systems, and use widgets to make actions directly executable inside the conversation. We help you set up agents, connections and widgets.
That doesn't mean you have to start big. You can begin with one clear use case: a frequently asked question that leads into a process, an appointment module driven from the conversation, or an order flow you start from the conversation. From there the functionality can be extended step by step.