For agents

Readable without a browser.

Everything on this site answers in text. Start from llms.txt, read any page as Markdown, or ask a page for its graph: what it shows, what can be done there, and where to go next. The graph is Affordances, and its actions run the same reference parser the range selector does, so an agent can test a range, desugar it or order versions by precedence with one request.

Try it

Three requests.

The graph of the range selector, one call to its select action, and the MCP server added to Claude Code. Each answers with JSON; the Markdown twins answer with text.

read a page's graph
curl -H 'Accept: application/vnd.affordances+json' https://semver.xyz/selector/
call an action
curl https://semver.xyz/selector/ \
  -H 'Content-Type: application/vnd.affordances+json' \
  -d '{"action":"select","input":{"range":"^1.0.0-rc","versions":["1.0.0-beta.2","1.0.0-rc.1","1.0.0","1.0.0+build.7"]}}'
connect an MCP client
claude mcp add --transport http semver https://semver.xyz/mcp

What an agent can rely on

Every action is public and readOnly: nothing here changes state, needs a session or asks for confirmation. The graphs and the Markdown are generated from the same sources as the pages (spec/SPEC.md, the extension Markdown, benchmarks/results), so they never disagree with what a person sees. Where a page and its Markdown could differ, the Markdown governs; the specification’s text is normative.