A short piece of fiction, written by API Evangelist. The reader is imagined; the post, and every number and quotation from it, is real: "Skills Over MCP Is Now Officially Beta," Kin Lane, October 29, 2026, https://apievangelist.com/2026/10/29/skills-over-mcp-is-now-officially-beta/
Marisol read the post on Thursday afternoon with a half-eaten sandwich beside her keyboard. She was a platform engineer at a company that made scheduling software for freight carriers, and for three weeks she had been circling a question her manager kept asking. Should their MCP server serve skills?
The first paragraph annoyed her. Pull request #161 had merged, the specification had moved from one folder to another, and the author said right away that nothing in the specification itself changed. She had come looking for news and found a rename. She almost closed the tab.
But the next part made her slow down. She had never noticed that the word stable, the one she had been reading as a promise, had only meant a released revision as opposed to a draft. Other extensions had used their own words for the same idea. She thought about the Slack message she had sent in September saying the skills spec was stable, and winced a little. She had meant it as reassurance. It had been closer to a guess.
The four stages helped. Experimental, Beta, Stable, Core. She copied them into her notes. Beta, as the post laid it out, meant that "breaking changes are expected to be rare" and that each one gets documented. That was weaker than what she had assumed, and it bothered her. Rare is not never. She had been hoping for a settled thing she could build once and forget, and the post was telling her, politely, that this was not that.
Still, the practical list was reassuring. "Existing implementations keep working." The identifier stayed the same and the skill:// resources stayed the same. Nothing she might have built in the last month needed to be torn out. She noticed the author was just as plain about the unknowns. The extensions page did not yet say whether Skills would end up as a Stable extension or be promoted into Core, and she felt some relief at being told that rather than having it papered over.
The number near the end stopped her. A week earlier the author had counted four companies serving skills over MCP the way the spec says, out of 1,450 public MCP endpoints that answered. Four. She felt two things at once. One was a small chill, because it meant almost nobody had gone first. The other was something closer to opportunity, because it meant the design was still soft enough to be shaped.
She was not fully persuaded by the cheerful ending. "Beta is an invitation, not a warning." That was easy to say if you were not the one explaining to a manager why a dependency could change. Invitations also come with costs. But she could not honestly argue that the label was wrong, only that it was less comfortable than the old one.
So she did something small. She did not write a proposal and she did not open a pull request. She opened the team's tracker and made a ticket asking someone to read the ext-skills decision log every other Friday, since the post said it now serves as the changelog for Beta breaking changes. She put her own name on it first. Then she added a note under the manager's question: yes, probably, but with a fortnightly look at the log, and no promises yet.
She finished the sandwich and went back to work.