Release v0.29.0

Your coding agent can now run the whole translation cycle. A new MCP server (@verbaly/mcp) gives Claude Code, Cursor and any MCP client four tools: see the coverage, list what's missing, extract new text and machine-translate the gaps, with every machine translation still landing as a draft a human reviews. A public Agent Skill and an llms.txt on the docs site complete the channel. Nuxt projects stop getting a verbaly.d.ts in their root (types now live in .nuxt/), and displayed code snippets in .vue/.astro/.svelte markup can no longer invent phantom keys. No breaking changes.

Highlights

• New: @verbaly/mcp, the translation cycle as agent tools. One command (claude mcp add verbaly -- npx -y @verbaly/mcp) and your coding agent can check coverage, list missing translations, extract new text and fill the gaps with machine translation. Anything the agent translates is saved as a draft, so nothing ships without a human saying yes.
• A public Agent Skill teaches your agent Verbaly. Install skills/verbaly from the repo into your project and the agent knows the write, extract, check, translate cycle and the rules that keep it safe (never hand-write keys, params must survive translation, empty means untranslated).
• The docs now speak LLM. The docs site publishes an llms.txt index, so agents that read documentation find their way around Verbaly without scraping HTML.
• Nuxt projects keep their root clean. The generated types now live inside .nuxt/ and register themselves automatically, exactly like the Astro integration already did. No more verbaly.d.ts at the project root, and no tsconfig edits either.
• Displayed code can no longer become a translation key. Docs and tutorial pages that show t`…` snippets as visible text (in .vue, .astro or .svelte files) used to risk those snippets being extracted as real keys. Now only code inside real expressions counts.

---

For the full details of this release, see the repository changelog.