A short piece of fiction, written by API Evangelist. The reader is imagined; the post, and every number and quotation from it, is real: "Booking.com Announces Its MCP Server In An A2A Card. How Expedia And Airbnb Compare.," Kin Lane, November 13, 2026, https://apievangelist.com/2026/11/13/booking-com-announces-its-mcp-server-in-an-a2a-card-how-expedia-and-airbnb-compare/
Dara maintained the developer portal for a small travel startup, and she read the post on her phone while the kettle boiled. It was the kind of Friday item she usually skimmed. This one had a table, so she stayed.
The post was about Booking.com. It serves an agent card at developers.booking.com, and the card calls itself "Developer Documentation". Dara noticed what the card didn't do. It didn't claim to be an agent. It announced an MCP server at developers.booking.com/mcp and left it there. The author's verdict was that "the agent card is a signpost". She liked that, and it was a little embarrassing how much she liked it. Her own team had spent two sprints arguing about what to put in a card, and here was a large company shipping one that was mostly a pointer.
Then she read the details and her mood cooled. The skills array was empty. The endpoint at /a2a answered a POST with a 405. The protocol version was 0.3.0, while the post says A2A is now at 1.0. The card's own version was 0.0.1. The catalog still graded it conformant, with three noted deviations. She wasn't sure what to make of a grade that passed something nobody could call. Conformant seemed to mean the file was shaped correctly. It didn't mean anyone could use it.
She also bristled at the Expedia and Airbnb sections. The author had probed four Expedia hosts and three Airbnb hosts, found 404s, and concluded there was no card. That was fair as far as it went. But the table put "No" beside two names, and a reader skimming it could take that for a verdict on strategy. A missing file doesn't show a missing plan. For all she knew, both companies had agents in the works and just hadn't published a card yet. The post never said otherwise, and it was careful about the Airbnb developer host, where both paths returned a 200 with an HTML page. The author didn't count that as a card. Dara respected that. It was exactly the kind of mistake her own monitoring would make.
Still, the closing advice stayed with her. The post said the cheapest first step is a card at the canonical well-known path on the developer host, and the better step is real skills behind a real endpoint. It said "the fix is small". She wanted that to be true of her own portal. She suspected it wasn't, because her team had never agreed on what a skill even was.
She finished the tea and did not open a planning document. She opened a terminal and ran a plain request against her own developer host at the well-known path, then the second path the post had tried. She wanted to see whether her site would return a 404 or a pleasant 200 with an HTML page.
It returned the HTML page.
Dara copied the response headers into a note, wrote one line saying the page was not a card and should not be counted as one, and put the note at the top of Monday's list. Then she closed the post. She still thought the Expedia table was a little harsh.