AI agents will soon be asking your website for a quote. We tried it on meom.fi
Key takeaways
- Buyers are increasingly handing their research to AI agents, which, for now, browse websites and guess.
- Since 23 October 2026, meom.fi has had a public MCP server where agents can look up our services, customer stories, facts about MEOM, team and booking link.
- The server runs in the same WordPress installation as the website, so an agent gets the same information as someone reading the site.
- The trial showed what an agent needs from a website: structured service information, measured results and consistent figures.
- We’re measuring usage for four weeks, and you can test the server yourself right now.
Imagine a marketing director typing into their AI assistant: “Find three Finnish B2B website agencies with a customer story from manufacturing, and book me an intro meeting.” The agent opens dozens of pages, picks up whatever it can find and guesses the rest. It infers each customer’s industry from the description, never finds the results and misses the booking link buried in the footer.
An agent doesn’t need to guess when the website answers it directly. That’s why we’ve opened up an API on meom.fi that agents can ask about the same things a buyer would otherwise look for on our website. This is a trial, not a product. Here’s what it already does, what it doesn’t do yet and what we’re measuring.
A buyer’s AI agent can now ask meom.fi directly
On 23 October 2026, we launched a public, read-only MCP server on meom.fi. MCP (Model Context Protocol) is an open standard that lets AI assistants such as Claude and ChatGPT connect to external tools and data sources. Once an assistant is connected to our server, it can use five tools to ask for:
- our services, how each one works and frequently asked questions
- customer stories, the results they report and what each customer says in their own words
- facts about MEOM, each with a source page and the date it was checked
- our team members’ names, roles and short bios
- a link a person can use to book a meeting.
We built the server with WordPress’s own tools (the Abilities API and the MCP Adapter), inside the same WordPress installation that runs the website. So an agent gets the same information as someone reading the site, and there’s no separate copy quietly going out of date. The server is separate from the internal MCP connection our team uses to create content for the site.
The trial gave us a long to-do list, and that’s good news
Gathering the information for the agent made us look at our website, for the first time, the way a buyer’s agent sees it. The information was there, but it was written for a person who reads a page from top to bottom. An agent needs it in a form it can pick out piece by piece. We came away with three big jobs:
- Only 7 of our 52 Finnish-language customer stories report a measured result. There’s a good project behind every one of the others, and now we know it’s worth digging out their results and writing them up.
- Our customer stories aren’t tagged by industry. Once we add that, the question “do you have experience in manufacturing?” gets a straight answer.
- Our customer satisfaction and ROI figures varied from page to page. We’re making them consistent, and then we’ll give them to the agent too.
This is exactly why we ran the trial. Without it, we’d have carried on developing the website with human readers alone in mind and noticed the gaps only once agents started asking. Now we have a list of content work that will make the site better for both: clearer for people and easier for agents to get answers from.
We do the same work with our customers in our ongoing generative engine optimisation (GEO) service. Every month, we find out whether AI is selling their company, and we fix whatever on the website is getting in the way.
Agents and AI search ask the same question: can this company be recommended, and why? When a buyer asks ChatGPT, Gemini or Copilot which agency has built a B2B website for a manufacturer, a mention alone doesn’t sell. AI should be able to say what you do, who for and with what results. We expect the same fixes to help there too, but we haven’t measured it yet. That’s why the trial is part of our GEO work, where we find out whether AI sells a company or merely knows it exists.
Three things you can check on your own website today
An API doesn’t have to come first. Agents benefit from the same fixes that help human readers and AI search:
- Include one measured result in every customer story. “Leads went up” isn’t enough for an agent; “we received twice as many requests for quotes as the year before” is.
- Tag your customer stories with industry and service. Then the question “do you have experience in our sector?” gets an answer without any guesswork.
- Go through the figures that appear on several pages. Your number of customers, founding year, satisfaction score and project duration should be stated the same way everywhere.
Once these are in order, an API is a small extra job. Without them, it just serves the agent the same muddled picture, only faster.
AI agents can’t find our MCP server on their own yet
To be honest, few agents ask meom.fi anything today unless someone has connected the server to them. Agents don’t yet go looking for MCP servers on websites unprompted. The user has to add the server themselves, or find it in a directory.
We’ve made the server as easy to find as we can for now. It’s described in the site’s llms.txt file and at /.well-known/mcp.json, and it’s listed in the official MCP Registry as fi.meom/public.
We’re measuring usage from 23 October to 20 November 2026. For every call, the server logs the time, the tool used and the search parameters passed to it, the name and version the assistant reports, and the application identifier (User-Agent). It doesn’t store IP addresses, and the logs are deleted after 60 days. They show us which assistants connect, what they ask and whether they request the booking link. We’ll share the results after the measurement period, even if there’s little to report. The value of the trial lies in what we learn and the evidence it gives us, not in buyer traffic. By the time agents start asking, we want to know what they’ll ask.
How to test the MCP server yourself
The MCP server’s address is https://www.meom.fi/wp-json/meom-public/v1/mcp. You don’t need an account or a key, and the server is read-only.
Claude (claude.ai or the desktop app)
Open Settings → Connectors → Add custom connector. Give it a name such as MEOM and paste in the address. Whether custom connectors are available depends on your Claude plan. Then try asking, for example:
- “Use the MEOM connector: does MEOM have WordPress customer stories from manufacturing, and what results do they report?”
- “Explain, with sources, why I should or shouldn’t choose MEOM for a HubSpot website redesign.”
- “How does a MEOM website project work, and how long does it take?”
- “I’d like to talk to MEOM about GEO. Where can I book a meeting?”
Claude Code
Add the server with this command:
claude mcp add --transport http meom https://www.meom.fi/wp-json/meom-public/v1/mcp
MCP Inspector
If you’re technically curious, you can look at the server’s tools in your browser. Run npx @modelcontextprotocol/inspector, choose Streamable HTTP as the Transport Type, paste the address into the URL field and click Connect. You’ll see the tools and can call them one by one.
Agent readiness starts where a good website does
An agent asks the same things a buyer does: what do you do, who have you done it for, why you, and when can we talk? If your website answers those questions clearly for people, you’re already halfway there. The rest is structure, consistent figures and an API that serves the answers up to agents.
Want to know what AI is already telling buyers about your company? Book a free 30-minute GEO assessment. You’ll get a summary of where you stand today and three concrete actions. If you’d like the same API on your own WordPress website, get in touch.