v0.2.0

Highlights

The server is now a pure Lucid bridge. It exports diagrams as images and lets your MCP client's own vision-capable model interpret them, so no AI provider keys are needed.

Removed

• Built-in image-analysis backend (Azure OpenAI / OpenAI). get-document with analyzeImage: true now returns the page title and the PNG image block only. The analyzeImage parameter name is kept for compatibility; it now just toggles PNG export.
• openai dependency and all AZURE_OPENAI_* / OPENAI_* environment variables.
• Smithery deployment config (smithery.yaml, Dockerfile).

Changed

• Interpreting diagrams now requires a vision-capable model in the MCP client. See "Client and Model Compatibility" in the README (verified with Claude Code, Codex CLI, and OpenCode with a vision model).
• Refactored the README install instructions to be client-agnostic, with a Claude Code example showing the only required setting, LUCID_API_KEY.

Fixed

• Build against @modelcontextprotocol/sdk 1.16+ (capabilities moved to the McpServer options argument).

Internal

• Node.js >= 22 baseline; CI on Node 22 and 24 with current GitHub Actions.
• Committed package-lock.json, CI uses npm ci, the build cleans its output directory, and publishing uses npm trusted publishing (OIDC) with automatic provenance.

Full changelog: see CHANGELOG.md.