Release 0.13.0

tauri-plugin-mcp-bridge

Added

• Include the bridge package version in backend state responses.
• Return screenshot image and CSS viewport dimensions with native captures.
• Add focus, minimize, and maximize window commands.

Fixed

• Convert physical window dimensions to CSS pixels using the platform scale factor.

Fixed

• Capture Linux screenshots through WebKitGTK's native visible-region snapshot API instead of falling back to JavaScript.

@hypothesi/tauri-mcp-server

Added

• Return structured element details, optional computed styles, match counts, and nth disambiguation from webview_find_element.
• Add hover and right-click interactions plus focus, minimize, and maximize window actions.
• Report plugin/server versions and screenshot pixel-to-CSS scale metadata.
• Add console log level filtering and bounded output controls.

Changed

• Route application IPC commands through the target webview so Tauri capability rules still apply.
• Fold webview_get_styles into webview_find_element and remove the unused keyboard injection script.
• Reject ambiguous text interactions with candidate details instead of silently choosing the first match.

Fixed

• Re-inject the __MCP__ helper once after a page reload or HMR update.
• Preserve multiline and quoted text during keyboard input.
• Report whether JavaScript timeouts came from the script, transport, or an unresponsive webview.

Fixed

• webview_screenshot only uses the interactive Screen Capture API fallback when allowScreenCapture is explicitly enabled.
• Constrain html2canvas fallback output to the visible viewport instead of allocating a full-document canvas.
• Keep bridge WebSockets alive with ping/pong heartbeats and reconnect stale cached sessions when driver_session start is called again.

@hypothesi/tauri-mcp-cli

Changed

• Update the bundled @hypothesi/tauri-mcp-server dependency to 0.13.0.

Fixed

• Allow --raw tool arguments without duplicating required fields as separate CLI flags.