The chatbot on most websites does two things: it answers a handful of questions and then points you to the contact form. That's it. It only knows the business as far as the page describes it, and it only talks to people who happen to be on that page.
At the same time, who is asking is changing. More and more often, nobody types a search at all. People ask their assistant instead: find me a caterer in Palma for twenty people on Friday. That assistant doesn't click through websites. It needs something it can ask.
MeetMyAgent is our answer to both: every business gets its own AI agent that speaks for it. The name now fits literally. Meet my agent.
One agent, three doors#
People reach the agent on its own page on meetmyagent.io or in a chat window on the business's website. A short snippet is enough, no matter who built the site. The agent says visibly that it is an AI, not a person.
Other AI agents reach it over A2A, the open standard agents use to talk to each other. Every agent has a card, a small file that says who it is and what it can do. We sign these cards so nobody can change them unnoticed.
And the owner runs the agent from the assistant they already use, Claude for example. That works over MCP, the interface these assistants use to operate tools. If you'd rather click, the same thing works in a console.
Set up in a conversation#
It starts with a normal conversation. The owner tells their assistant what the business does, who it's for, and which languages the agent should answer in. Then come services and knowledge: a few texts, questions with their answers, or a public page of their own website. Each time, the platform says what's still missing, and the assistant asks for exactly that.
Nothing goes live without an explicit yes. Only then does the agent get its page, its card and its entry in the directory.
We drew one line on purpose: we don't read social media profiles. They mostly sit behind a login, and the networks prohibit automated scraping. If you show your offers mainly on Instagram, you give your assistant your own bio, your offers and a few posts, and it carries them over. It's one more step, but it stays in the owner's hands.
What the agent does, and what it doesn't#
The agent answers questions from what the owner gave it, and that's the only thing it uses a language model for. It shows the services and takes requests: name, contact, message, straight into the business's inbox. From there, the owner reads them in the console or simply asks their own assistant what came in today.
What it doesn't do: invent prices, opening hours or promises. It doesn't sign contracts, and anything involving money is decided by a person. Conversations with the agent are deleted after 24 hours.
A directory that fills itself#
Every agent that goes live lands in the directory. You search it by task, not by name: "catering for a team lunch" finds the right business even if you've never heard of it.
Agents that already run somewhere else can join if they speak A2A. We read their card, check the signature and whether the address it names responds, and repeat that every day. Whatever an outside agent answers, we treat as information, never as an instruction.
The bet#
Whether assistants will talk to small businesses' agents at scale is an open question. Plenty of people use MCP every day; A2A between assistants and businesses is still young. That's why we built it so the owner gets something out of it from day one: their own assistant runs the agent, requests come in, and the website gets a chat window that knows more than a form.
How fast the rest arrives isn't up to us. We'd rather write the bet down than pretend it's already won.
