@hypothesi/tauri-mcp-server 0.10.0 → 0.11.0

package/dist/driver/element-picker.js

@@ -99,13 +99,13 @@ async function cleanupPickerHighlights(windowId, appIdentifier) {
  * Capture a screenshot of a specific element using html2canvas.
  * Returns the base64 data URL of the cropped element image, or null on failure.
  */
-async function captureElementScreenshot(cssSelector, windowId) {
+async function captureElementScreenshot(cssSelector, windowId, appIdentifier) {
     // Ensure html2canvas is loaded in the webview
     try {
-        const isRegistered = await isScriptRegistered(HTML2CANVAS_SCRIPT_ID);
+        const isRegistered = await isScriptRegistered(HTML2CANVAS_SCRIPT_ID, appIdentifier);
         if (!isRegistered) {
             const source = getHtml2CanvasSource();
-            await registerScript(HTML2CANVAS_SCRIPT_ID, 'inline', source);
+            await registerScript(HTML2CANVAS_SCRIPT_ID, 'inline', source, windowId, appIdentifier);
         }
     }
     catch {
@@ -149,7 +149,7 @@ async function captureElementScreenshot(cssSelector, windowId) {
       return dataUrl;
    `;
     try {
-        const dataUrl = await executeAsyncInWebview(captureScript, windowId, 10000);
+        const dataUrl = await executeAsyncInWebview(captureScript, windowId, 10000, appIdentifier);
         if (!dataUrl || !dataUrl.startsWith('data:image/')) {
             return null;
         }
@@ -221,7 +221,7 @@ export async function selectElement(options) {
     // Add formatted metadata
     content.push({ type: 'text', text: formatElementMetadata(element) });
     // Capture element-only screenshot (no picker overlays visible)
-    const screenshot = await captureElementScreenshot(element.cssSelector, windowId);
+    const screenshot = await captureElementScreenshot(element.cssSelector, windowId, appIdentifier);
     if (screenshot) {
         content.push(screenshot);
     }
@@ -261,7 +261,7 @@ export async function getPointedElement(options) {
     // Add formatted metadata
     content.push({ type: 'text', text: formatElementMetadata(element) });
     // Capture element-only screenshot (no overlays)
-    const screenshot = await captureElementScreenshot(element.cssSelector, windowId);
+    const screenshot = await captureElementScreenshot(element.cssSelector, windowId, appIdentifier);
     if (screenshot) {
         content.push(screenshot);
     }

package/dist/driver/script-manager.d.ts

@@ -59,7 +59,7 @@ interface GetScriptsResponse {
  * @param windowLabel - Optional window label to target
  * @returns Promise resolving to registration result
  */
-export declare function registerScript(id: string, type: ScriptType, content: string, windowLabel?: string): Promise<RegisterScriptResponse>;
+export declare function registerScript(id: string, type: ScriptType, content: string, windowLabel?: string, appIdentifier?: string | number): Promise<RegisterScriptResponse>;
 /**
  * Removes a script from the registry and DOM.
  *
@@ -67,25 +67,25 @@ export declare function registerScript(id: string, type: ScriptType, content: st
  * @param windowLabel - Optional window label to target
  * @returns Promise resolving to removal result
  */
-export declare function removeScript(id: string, windowLabel?: string): Promise<RemoveScriptResponse>;
+export declare function removeScript(id: string, windowLabel?: string, appIdentifier?: string | number): Promise<RemoveScriptResponse>;
 /**
  * Clears all registered scripts from the registry and DOM.
  *
  * @param windowLabel - Optional window label to target
  * @returns Promise resolving to the number of scripts cleared
  */
-export declare function clearScripts(windowLabel?: string): Promise<ClearScriptsResponse>;
+export declare function clearScripts(windowLabel?: string, appIdentifier?: string | number): Promise<ClearScriptsResponse>;
 /**
  * Gets all registered scripts.
  *
  * @returns Promise resolving to the list of registered scripts
  */
-export declare function getScripts(): Promise<GetScriptsResponse>;
+export declare function getScripts(appIdentifier?: string | number): Promise<GetScriptsResponse>;
 /**
  * Checks if a script with the given ID is registered.
  *
  * @param id - The script ID to check
  * @returns Promise resolving to true if the script is registered
  */
-export declare function isScriptRegistered(id: string): Promise<boolean>;
+export declare function isScriptRegistered(id: string, appIdentifier?: string | number): Promise<boolean>;
 export {};

package/dist/driver/script-manager.js

@@ -19,8 +19,8 @@ import { ensureSessionAndConnect } from './plugin-client.js';
  * @param windowLabel - Optional window label to target
  * @returns Promise resolving to registration result
  */
-export async function registerScript(id, type, content, windowLabel) {
-    const client = await ensureSessionAndConnect();
+export async function registerScript(id, type, content, windowLabel, appIdentifier) {
+    const client = await ensureSessionAndConnect(appIdentifier);
     const response = await client.sendCommand({
         command: 'register_script',
         args: { id, type, content, windowLabel },
@@ -37,8 +37,8 @@ export async function registerScript(id, type, content, windowLabel) {
  * @param windowLabel - Optional window label to target
  * @returns Promise resolving to removal result
  */
-export async function removeScript(id, windowLabel) {
-    const client = await ensureSessionAndConnect();
+export async function removeScript(id, windowLabel, appIdentifier) {
+    const client = await ensureSessionAndConnect(appIdentifier);
     const response = await client.sendCommand({
         command: 'remove_script',
         args: { id, windowLabel },
@@ -54,8 +54,8 @@ export async function removeScript(id, windowLabel) {
  * @param windowLabel - Optional window label to target
  * @returns Promise resolving to the number of scripts cleared
  */
-export async function clearScripts(windowLabel) {
-    const client = await ensureSessionAndConnect();
+export async function clearScripts(windowLabel, appIdentifier) {
+    const client = await ensureSessionAndConnect(appIdentifier);
     const response = await client.sendCommand({
         command: 'clear_scripts',
         args: { windowLabel },
@@ -70,8 +70,8 @@ export async function clearScripts(windowLabel) {
  *
  * @returns Promise resolving to the list of registered scripts
  */
-export async function getScripts() {
-    const client = await ensureSessionAndConnect();
+export async function getScripts(appIdentifier) {
+    const client = await ensureSessionAndConnect(appIdentifier);
     const response = await client.sendCommand({
         command: 'get_scripts',
         args: {},
@@ -87,7 +87,7 @@ export async function getScripts() {
  * @param id - The script ID to check
  * @returns Promise resolving to true if the script is registered
  */
-export async function isScriptRegistered(id) {
-    const { scripts } = await getScripts();
+export async function isScriptRegistered(id, appIdentifier) {
+    const { scripts } = await getScripts(appIdentifier);
     return scripts.some((s) => { return s.id === id; });
 }

package/dist/driver/scripts/index.js

@@ -56,8 +56,24 @@ export function buildTypeScript(selector, text, strategy) {
          if (!element) throw new Error('Element not found: ' + selector);
 
          element.focus();
-         element.value = text;
-         element.dispatchEvent(new Event('input', { bubbles: true }));
+
+         // Use native prototype setter to bypass React's value tracker
+         var proto = element.tagName === 'TEXTAREA'
+            ? HTMLTextAreaElement.prototype
+            : HTMLInputElement.prototype;
+         var descriptor = Object.getOwnPropertyDescriptor(proto, 'value');
+
+         if (descriptor && descriptor.set) {
+            descriptor.set.call(element, text);
+         } else {
+            element.value = text;
+         }
+
+         // Reset React's internal value tracker so it detects the change
+         if (element._valueTracker) element._valueTracker.setValue('');
+
+         // Dispatch proper InputEvent (not generic Event) for React compatibility
+         element.dispatchEvent(new InputEvent('input', { bubbles: true, inputType: 'insertText', data: text }));
          element.dispatchEvent(new Event('change', { bubbles: true }));
 
          var msg = 'Typed "' + text + '" into ' + selector;

package/dist/driver/scripts/resolve-ref.js

@@ -47,9 +47,22 @@
       }
 
       if (strategy === 'text') {
+         // First try: match element text content
          var xpath = xpathForText(selectorOrRef);
          var result = document.evaluate(xpath, document, null, XPathResult.FIRST_ORDERED_NODE_TYPE, null);
-         return result.singleNodeValue;
+         if (result.singleNodeValue) return result.singleNodeValue;
+
+         // Fallback: search placeholder, aria-label, and title attributes
+         var attrSelectors = [
+            '[placeholder*="' + selectorOrRef.replace(/"/g, '\\"') + '"]',
+            '[aria-label*="' + selectorOrRef.replace(/"/g, '\\"') + '"]',
+            '[title*="' + selectorOrRef.replace(/"/g, '\\"') + '"]',
+         ];
+         for (var i = 0; i < attrSelectors.length; i++) {
+            var el = document.querySelector(attrSelectors[i]);
+            if (el) return el;
+         }
+         return null;
       }
 
       if (strategy === 'xpath') {
@@ -78,12 +91,25 @@
       }
 
       if (strategy === 'text') {
+         // First try: match element text content
          var xpath = xpathForText(selector);
          var snapshot = document.evaluate(xpath, document, null, XPathResult.ORDERED_NODE_SNAPSHOT_TYPE, null);
          var results = [];
          for (var i = 0; i < snapshot.snapshotLength; i++) {
             results.push(snapshot.snapshotItem(i));
          }
+         if (results.length > 0) return results;
+
+         // Fallback: search placeholder, aria-label, and title attributes
+         var attrSelectors = [
+            '[placeholder*="' + selector.replace(/"/g, '\\"') + '"]',
+            '[aria-label*="' + selector.replace(/"/g, '\\"') + '"]',
+            '[title*="' + selector.replace(/"/g, '\\"') + '"]',
+         ];
+         for (var i = 0; i < attrSelectors.length; i++) {
+            var found = Array.from(document.querySelectorAll(attrSelectors[i]));
+            if (found.length > 0) return results.concat(found);
+         }
          return results;
       }
 

package/dist/driver/webview-executor.d.ts

@@ -12,7 +12,7 @@ import { z } from 'zod';
  *
  * @throws Error if no session is active (driver_session must be called first)
  */
-export declare function ensureReady(): Promise<void>;
+export declare function ensureReady(windowId?: string, appIdentifier?: string | number): Promise<void>;
 /**
  * Reset initialization state (useful for testing or reconnecting).
  */
@@ -48,7 +48,7 @@ export declare function executeInWebviewWithContext(script: string, windowId?: s
  * @param timeout - Timeout in milliseconds (default: 5000)
  * @returns Result of the script execution
  */
-export declare function executeAsyncInWebview(script: string, windowId?: string, timeout?: number): Promise<string>;
+export declare function executeAsyncInWebview(script: string, windowId?: string, timeout?: number, appIdentifier?: string | number): Promise<string>;
 /**
  * Initialize console log capture in the webview.
  * This intercepts console methods and stores logs in memory.

package/dist/driver/webview-executor.js

@@ -1,6 +1,5 @@
 import { z } from 'zod';
-import { connectPlugin } from './plugin-client.js';
-import { hasActiveSession, getDefaultSession, resolveTargetApp } from './session-manager.js';
+import { hasActiveSession, resolveTargetApp, manageDriverSession } from './session-manager.js';
 import { createMcpLogger } from '../logger.js';
 import { buildScreenshotScript, buildScreenshotCaptureScript, getHtml2CanvasSource, HTML2CANVAS_SCRIPT_ID, } from './scripts/html2canvas-loader.js';
 import { registerScript, isScriptRegistered } from './script-manager.js';
@@ -17,7 +16,7 @@ import { getResolveRefSource, RESOLVE_REF_SCRIPT_ID } from './scripts/index.js';
 // ============================================================================
 // Auto-Initialization System
 // ============================================================================
-let isInitialized = false;
+const initializedTargets = new Set();
 const driverLogger = createMcpLogger('DRIVER');
 /**
  * Ensures the MCP server is fully initialized and ready to use.
@@ -32,28 +31,52 @@ const driverLogger = createMcpLogger('DRIVER');
  *
  * @throws Error if no session is active (driver_session must be called first)
  */
-export async function ensureReady() {
-    if (isInitialized) {
-        return;
-    }
-    // Require an active session to prevent connecting to wrong app
+export async function ensureReady(windowId, appIdentifier) {
+    // Auto-connect if no active session
     if (!hasActiveSession()) {
-        throw new Error('No active session. Call driver_session with action "start" first to connect to a Tauri app.');
+        const result = await manageDriverSession('start');
+        if (!hasActiveSession()) {
+            throw new Error('Auto-connect failed: ' + result + '. Call driver_session with action "start" to connect manually.');
+        }
     }
-    // Get default session for initial connection
-    const session = getDefaultSession();
-    if (session) {
-        await connectPlugin(session.host, session.port);
+    const session = resolveTargetApp(appIdentifier);
+    if (!session.client.isConnected()) {
+        await session.client.connect();
     }
-    // Register the resolve-ref helper so ref-based selectors work in all tools
-    await registerScript(RESOLVE_REF_SCRIPT_ID, 'inline', getResolveRefSource());
-    isInitialized = true;
+    const targetKey = `${session.host}:${session.port}:${windowId ?? 'main'}`;
+    if (initializedTargets.has(targetKey)) {
+        return;
+    }
+    // Register the resolve-ref helper in the target window
+    // so ref-based selectors work there.
+    await registerScript(RESOLVE_REF_SCRIPT_ID, 'inline', getResolveRefSource(), windowId, appIdentifier);
+    await waitForResolveRefHelper(session, windowId);
+    initializedTargets.add(targetKey);
 }
 /**
  * Reset initialization state (useful for testing or reconnecting).
  */
 export function resetInitialization() {
-    isInitialized = false;
+    initializedTargets.clear();
+}
+async function waitForResolveRefHelper(session, windowId) {
+    if (!session) {
+        throw new Error('No active session available while registering resolve-ref helper.');
+    }
+    for (let attempt = 0; attempt < 20; attempt++) {
+        const response = await session.client.sendCommand({
+            command: 'execute_js',
+            args: {
+                script: 'return !!(window.__MCP__ && typeof window.__MCP__.resolveRef === "function")',
+                windowLabel: windowId,
+            },
+        }, 2000);
+        if (response.success && response.data === true) {
+            return;
+        }
+        await new Promise((resolve) => { return setTimeout(resolve, 50); });
+    }
+    throw new Error('Resolve-ref helper was not available in the webview after registration.');
 }
 /**
  * Execute JavaScript in the Tauri webview using native IPC via WebSocket.
@@ -78,7 +101,7 @@ export async function executeInWebview(script, windowId, appIdentifier) {
 export async function executeInWebviewWithContext(script, windowId, appIdentifier) {
     try {
         // Ensure we're fully initialized
-        await ensureReady();
+        await ensureReady(windowId, appIdentifier);
         // Resolve target session
         const session = resolveTargetApp(appIdentifier);
         const client = session.client;
@@ -124,11 +147,12 @@ export async function executeInWebviewWithContext(script, windowId, appIdentifie
  * @param timeout - Timeout in milliseconds (default: 5000)
  * @returns Result of the script execution
  */
-export async function executeAsyncInWebview(script, windowId, timeout = 5000) {
+export async function executeAsyncInWebview(script, windowId, timeout, appIdentifier) {
+    const resolvedTimeout = timeout ?? 5000;
     const wrappedScript = `
       return (async () => {
          const timeoutPromise = new Promise((_, reject) => {
-            setTimeout(() => reject(new Error('Script execution timeout')), ${timeout});
+            setTimeout(() => reject(new Error('Script execution timeout')), ${resolvedTimeout});
          });
 
          const scriptPromise = (async () => {
@@ -138,7 +162,7 @@ export async function executeAsyncInWebview(script, windowId, timeout = 5000) {
          return await Promise.race([scriptPromise, timeoutPromise]);
       })();
    `;
-    return executeInWebview(wrappedScript, windowId);
+    return executeInWebview(wrappedScript, windowId, appIdentifier);
 }
 // ============================================================================
 // Console Log Capture System
@@ -268,14 +292,14 @@ function buildScreenshotResult(dataUrl, method, windowContext) {
  * Prepares the html2canvas script for screenshot capture.
  * Tries to use the script manager for persistence, falls back to inline injection.
  */
-async function prepareHtml2canvasScript(format, quality) {
+async function prepareHtml2canvasScript(format, quality, windowId, appIdentifier) {
     try {
         // Check if html2canvas is already registered
-        const isRegistered = await isScriptRegistered(HTML2CANVAS_SCRIPT_ID);
+        const isRegistered = await isScriptRegistered(HTML2CANVAS_SCRIPT_ID, appIdentifier);
         if (!isRegistered) {
             // Register html2canvas via script manager for persistence across navigations
             const html2canvasSource = getHtml2CanvasSource();
-            await registerScript(HTML2CANVAS_SCRIPT_ID, 'inline', html2canvasSource);
+            await registerScript(HTML2CANVAS_SCRIPT_ID, 'inline', html2canvasSource, windowId, appIdentifier);
         }
         // Use the capture-only script since html2canvas is now registered
         return buildScreenshotCaptureScript(format, quality);
@@ -299,7 +323,7 @@ export async function captureScreenshot(options = {}) {
     // - Linux: Chromium/WebKit screenshot APIs
     try {
         // Ensure we're fully initialized
-        await ensureReady();
+        await ensureReady(windowId, appIdentifier);
         // Resolve target session
         const session = resolveTargetApp(appIdentifier);
         const client = session.client;
@@ -331,7 +355,7 @@ export async function captureScreenshot(options = {}) {
     }
     // Fallback 1: Use html2canvas library for high-quality DOM rendering
     // Try to use the script manager to register html2canvas for persistence
-    const html2canvasScript = await prepareHtml2canvasScript(format, quality);
+    const html2canvasScript = await prepareHtml2canvasScript(format, quality, windowId, appIdentifier);
     // Fallback: Try Screen Capture API if available
     // Note: This script is wrapped by executeAsyncInWebview, so we don't need an IIFE
     const screenCaptureScript = `
@@ -390,7 +414,7 @@ export async function captureScreenshot(options = {}) {
    `;
     try {
         // Try html2canvas second (after native APIs)
-        const result = await executeAsyncInWebview(html2canvasScript, undefined, 10000); // Longer timeout for library loading
+        const result = await executeAsyncInWebview(html2canvasScript, windowId, 10000, appIdentifier);
         // Validate that we got a real data URL, not 'null' or empty
         if (result && result !== 'null' && result.startsWith('data:image/')) {
             return buildScreenshotResult(result, 'html2canvas');
@@ -400,7 +424,7 @@ export async function captureScreenshot(options = {}) {
     catch (html2canvasError) {
         try {
             // Fallback to Screen Capture API
-            const result = await executeAsyncInWebview(screenCaptureScript);
+            const result = await executeAsyncInWebview(screenCaptureScript, windowId, 5000, appIdentifier);
             // Validate that we got a real data URL
             if (result && result.startsWith('data:image/')) {
                 return buildScreenshotResult(result, 'Screen Capture API');

package/dist/driver/webview-interactions.js

@@ -22,7 +22,8 @@ export const WindowTargetSchema = z.object({
  * Defaults to 'css' for backward compatibility.
  */
 const selectorStrategyField = z.enum(['css', 'xpath', 'text']).default('css').describe('Selector strategy: "css" (default) for CSS selectors, "xpath" for XPath expressions, ' +
-    '"text" to find elements containing the given text. Ref IDs (e.g., "ref=e3") work with any strategy.');
+    '"text" to find elements by text content, with fallback to placeholder, aria-label, ' +
+    'and title attributes. Ref IDs (e.g., "ref=e3") work with any strategy.');
 // ============================================================================
 // Schemas
 // ============================================================================
@@ -154,10 +155,8 @@ export async function screenshot(options = {}) {
         if (!imageContent || imageContent.type !== 'image') {
             throw new Error('Screenshot capture failed: no image data');
         }
-        // Decode base64 and write to file
-        const buffer = Buffer.from(imageContent.data, 'base64');
         const resolvedPath = resolve(filePath);
-        await writeFile(resolvedPath, buffer);
+        await writeFile(resolvedPath, imageContent.data, 'base64');
         return { filePath: resolvedPath, format };
     }
     return result;
@@ -295,7 +294,7 @@ export async function domSnapshot(options) {
     const { type, selector, strategy, windowId, appIdentifier } = options;
     // Only load aria-api for accessibility snapshots
     if (type === 'accessibility') {
-        await ensureAriaApiLoaded(windowId);
+        await ensureAriaApiLoaded(windowId, appIdentifier);
     }
     // Then execute the snapshot script
     const script = buildScript(SCRIPTS.domSnapshot, { type, selector: selector ?? null, strategy: strategy ?? 'css' });
@@ -311,11 +310,11 @@ export async function domSnapshot(options) {
  * Ensure aria-api library is loaded in the webview.
  * Uses the script manager to inject the library if not already present.
  */
-async function ensureAriaApiLoaded(windowId) {
+async function ensureAriaApiLoaded(windowId, appIdentifier) {
     const { getAriaApiSource, ARIA_API_SCRIPT_ID: ariaApiScriptId } = await import('./scripts/aria-api-loader.js');
     const { registerScript, isScriptRegistered } = await import('./script-manager.js');
-    if (await isScriptRegistered(ariaApiScriptId)) {
+    if (await isScriptRegistered(ariaApiScriptId, appIdentifier)) {
         return;
     }
-    await registerScript(ariaApiScriptId, 'inline', getAriaApiSource(), windowId);
+    await registerScript(ariaApiScriptId, 'inline', getAriaApiSource(), windowId, appIdentifier);
 }

package/dist/tools-registry.js

@@ -171,6 +171,7 @@ export const TOOLS = [
         name: 'webview_find_element',
         description: '[Tauri Apps Only] Find DOM elements in a running Tauri app\'s webview. ' +
             'Supports CSS selectors (default), XPath expressions, and text content matching via the strategy parameter. ' +
+            'The "text" strategy first searches element text content, then falls back to placeholder, aria-label, and title attributes. ' +
             'Returns the element\'s HTML. ' +
             'Requires active driver_session. ' +
             MULTI_APP_DESC + ' ' +

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hypothesi/tauri-mcp-server",
-  "version": "0.10.0",
+  "version": "0.11.0",
   "mcpName": "io.github.hypothesi/mcp-server-tauri",
   "description": "A Model Context Protocol server for use with Tauri v2 applications",
   "type": "module",