@privacyscrubber/mcp-server 2.2.0 → 2.2.2

package/scrubber-core.cjs

@@ -66,7 +66,7 @@ function scrubText(text, customRules = [], tokenLabelMap = {}, profile = 'Genera
     
     // Protect system prompt from being scrubbed or counted
     let extractedSystemPrompt = "";
-    let textToProcess = text.replace(/[\u200b\u200c\u200d\ufeff]/g, '');
+    let textToProcess = text.replace(/[\u200b\u200c\u200d\ufeff\u00ad]/gi, '');
 
     const isSpecialized = profile && profile.toLowerCase() !== 'general';
     
@@ -119,6 +119,32 @@ function scrubText(text, customRules = [], tokenLabelMap = {}, profile = 'Genera
         });
     }
 
+    // Pre-scan input text for pre-existing tokens ([TYPE_N], <TYPE_N>, {{TYPE_N}}, __TYPE_N__)
+    // to advance counters and avoid counter collisions or overwriting literal quoted tokens
+    if (typeof textToProcess === 'string' && textToProcess) {
+        const existingTokenRegex = /(?:\[|<|\{\{|__)([A-Za-z0-9_]+)_(\d+)(?:\]|>|\}\}|__)/g;
+        let tMatch;
+        while ((tMatch = existingTokenRegex.exec(textToProcess)) !== null) {
+            const rawLabel = tMatch[1].toUpperCase();
+            const idx = parseInt(tMatch[2], 10);
+            if (!isNaN(idx)) {
+                let mappedType = rawLabel;
+                for (const [baseType, customLabel] of Object.entries(tokenLabelMap || {})) {
+                    if (String(customLabel).replace(/[^A-Za-z0-9_]/g, '_').toUpperCase() === rawLabel) {
+                        mappedType = baseType;
+                        break;
+                    }
+                }
+                mappedType = resolveTypeFromTokenLabel(mappedType);
+                if (counters[mappedType] !== undefined) {
+                    counters[mappedType] = Math.max(counters[mappedType], idx);
+                } else {
+                    customCounters[mappedType] = Math.max(customCounters[mappedType] || 0, idx);
+                }
+            }
+        }
+    }
+
     
     // Default labels
     const labels = {
@@ -289,6 +315,15 @@ function getLabelAliases(label) {
     return Array.from(aliases);
 }
 
+function resolveTypeFromTokenLabel(label) {
+    if (!label) return 'CUSTOM';
+    const upper = String(label).toUpperCase();
+    for (const [type, aliases] of Object.entries(LABEL_ALIASES)) {
+        if (aliases.includes(upper)) return type;
+    }
+    return upper;
+}
+
 function formatToken(label, index, format = 'brackets') {
     const cleanLabel = String(label || 'PII').replace(/[^A-Za-z0-9_]/g, '_').toUpperCase();
     switch(format) {
@@ -1090,6 +1125,16 @@ function hydrateRegex(rule) {
             }
         }
 
+        // Fuse adjacent text nodes that were split across chunk boundaries during LLM streaming mutations
+        try {
+            const assistantContainers = document.querySelectorAll(ASSISTANT_SELECTORS);
+            for (let i = 0; i < assistantContainers.length; i++) {
+                if (typeof assistantContainers[i].normalize === 'function') {
+                    assistantContainers[i].normalize();
+                }
+            }
+        } catch (_) {}
+
         walkTextNodes(document.documentElement);
 
         // TOGGLE RE-MASK: If no unmasked tokens were found on screen, but there are already revealed .ps-restored-data spans, re-mask them back to tokens

package/server.json

@@ -6,12 +6,12 @@
     "url": "https://github.com/moxno/privacyscrubber-mcp",
     "source": "github"
   },
-  "version": "2.2.0",
+  "version": "2.2.2",
   "packages": [
     {
       "registryType": "npm",
       "identifier": "@privacyscrubber/mcp-server",
-      "version": "2.2.0",
+      "version": "2.2.2",
       "transport": {
         "type": "stdio"
       },