v5.2.0

Added

• The publish workflow fails before npm publish when the GitHub Release tag
  and package.json version disagree. check-versions.mjs proved that
  package.json, server.json and the lockfile agreed with each other, but
  nothing tied that version to the tag the release was cut from — so tagging
  v4.9.0 on a commit reading 5.0.0 would have published 5.0.0 to npm and the
  MCP Registry while the GitHub Release — the artifact humans read — claimed
  otherwise, silently and on the irreversible side of the publish (#103).

Changed

• Raised the @modelcontextprotocol/sdk floor to ^1.30.0, which declares
  @hono/node-server: ^1.19.9 || ^2.0.5 where 1.29.0 declared only ^1.19.9.
  This does not change what a fresh, standalone consumer resolves — measured
  against a packed tarball in an empty project, both the already-published 5.1.0
  and this build resolve @hono/node-server 2.1.1 with npm audit --omit=dev
  clean, because ^1.29.0 already admitted SDK 1.30.0. What it buys is that this
  repo's @hono/node-server override no longer forces a major outside the
  SDK's declared range, and that a resolver cannot fall back to an SDK whose
  range predates the widening. The override itself stays as a floor pin: the
  SDK's ^2.0.5 still admits 2.0.5–2.0.9, which sit inside GHSA-9mqv-5hh9-4cgg
  (#81).
• The publish job runs on Node 24 (Active LTS) instead of Node 20. npm's Trusted
  Publishing prerequisite is two-part — npm 11.5.1+ and Node 22.14+ — and
  the job satisfied only the npm half, leaving the irreversible npm publish
  step one npm patch away from breaking if that floor starts being enforced. The
  CI test matrix stays on Node 20 + 22: it tracks engines.node, which is
  unchanged (#102).

Fixed

• Raised the brace-expansion override floor to ^5.0.9. GHSA-rgw5-rvv9-x895
  (published 2026-08-03) covers >= 4.0.0, < 5.0.9, so the previous ^5.0.8
  pin — correct when written — had drifted inside a live advisory range. This is
  a devDependency (eslint → minimatch) and never reached the --omit=dev CI
  gate, which was green throughout; the published runtime package is unaffected.
  It is the second rot on this same pin, and the guard test cannot catch the
  class: it asserts the floor it is given, so a range growing past a correct pin
  still produces a green run. That is now recorded on the entry itself (#112).