mcp-browser-extension 0.9.26 → 0.9.27

package/dist/src/executor/types.d.ts

@@ -310,6 +310,7 @@ export interface Executor {
         outer?: boolean;
     } & FrameOpts): Promise<{
         html: string;
+        frameId?: number;
     }>;
     /** Accessibility snapshot: interactive/landmark elements with stable refs the model can target. */
     snapshot(opts?: {

package/dist/src/mcp/helpers.d.ts

@@ -5,7 +5,7 @@
  * when the backend can (else sequences fill+click).
  * (Only `download_file` is privileged and lives on the executor.)
  */
-import type { Executor, FrameOpts } from '../executor/types';
+import { type Executor, type FrameOpts } from '../executor/types';
 export interface LinkOut {
     href: string;
     text: string;

package/dist/src/mcp/helpers.js

@@ -10,6 +10,8 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.extractLinks = extractLinks;
 exports.readAsMarkdown = readAsMarkdown;
 exports.fillForm = fillForm;
+const htmlparser2_1 = require("htmlparser2");
+const types_1 = require("../executor/types");
 const markdown_extract_1 = require("./markdown-extract");
 /**
  * Collect anchors from the page (or a subtree). Implemented as a single page
@@ -25,31 +27,99 @@ async function extractLinks(ex, args) {
     const root = ${root} ? document.querySelector(${root}) : document;
     if (!root) return [];
     const here = location.origin;
-    return [...root.querySelectorAll('a[href]')].map(a => ({
-      href: a.href, text: (a.textContent || '').trim().slice(0, 200),
-    })).filter(l => l.href && (${args.sameOriginOnly ? 'l.href.startsWith(here)' : 'true'}));
+    return [...root.querySelectorAll('a[href]')].filter(a => {
+      const raw = (a.getAttribute('href') || '').trim();
+      if (!raw || raw.startsWith('#')) return false;
+      try {
+        const url = new URL(a.href);
+        return !['javascript:', 'data:', 'blob:'].includes(url.protocol) &&
+          (${args.sameOriginOnly ? 'url.origin === here' : 'true'});
+      } catch { return false; }
+    }).map(a => ({ href: a.href, text: (a.textContent || '').trim().slice(0, 200) }));
   })()`;
     const frames = { frameId: args.frameId, allFrames: args.allFrames };
     let links;
-    const res = await ex.eval(expr, { tabId: args.tabId, ...frames });
-    if (res.ok && Array.isArray(res.value)) {
-        links = res.value;
+    try {
+        const res = await ex.eval(expr, { tabId: args.tabId, ...frames });
+        if (res.ok && Array.isArray(res.value))
+            links = res.value;
     }
-    else {
+    catch (err) {
+        if (!(err instanceof types_1.ExecutorError) || err.code !== 'POLICY_DENIED')
+            throw err;
+    }
+    if (!links) {
         // Fallback: parse hrefs out of the HTML (e.g. when eval is policy-denied).
-        const { html } = await ex.getHtml(args.selector ? { selector: args.selector } : undefined, {
+        const result = await ex.getHtml(args.selector ? { selector: args.selector } : undefined, {
             tabId: args.tabId,
             ...frames,
         });
-        links = [];
-        const re = /<a[^>]*href=["']([^"']+)["'][^>]*>([\s\S]*?)<\/a>/gi;
-        let m;
-        while ((m = re.exec(html)) !== null) {
-            links.push({ href: m[1], text: m[2].replace(/<[^>]+>/g, '').trim().slice(0, 200) });
+        const tabs = await ex.tabsList();
+        const tab = args.tabId ? tabs.find(t => t.tabId === args.tabId) : tabs.find(t => t.active);
+        let pageUrl = tab?.url;
+        const frameId = result.frameId ?? args.frameId;
+        if (frameId !== undefined && frameId !== 0) {
+            pageUrl = (await ex.framesList?.({ tabId: args.tabId }))?.find(f => f.frameId === frameId)?.url;
         }
+        if (!pageUrl)
+            throw new types_1.ExecutorError('TARGET_GONE', 'could not resolve the page URL for link extraction');
+        links = linksFromHtml(result.html, pageUrl, args.sameOriginOnly);
     }
     return { links: refineLinks(links, args) };
 }
+/** Parse serialized HTML without executing page code; the parser decodes entities. */
+function linksFromHtml(html, pageUrl, sameOriginOnly = false) {
+    const raw = [];
+    let current;
+    let baseHref;
+    let skippedDepth = 0;
+    new htmlparser2_1.Parser({
+        onopentag(name, attrs) {
+            if (skippedDepth || ['script', 'style', 'template'].includes(name)) {
+                skippedDepth++;
+                return;
+            }
+            if (name === 'base' && baseHref === undefined && attrs.href !== undefined)
+                baseHref = attrs.href;
+            if (name === 'a' && attrs.href !== undefined) {
+                current = { href: attrs.href.trim(), text: '' };
+                raw.push(current);
+            }
+        },
+        ontext(text) { if (current && !skippedDepth)
+            current.text += text; },
+        onclosetag(name) {
+            if (skippedDepth) {
+                skippedDepth--;
+                return;
+            }
+            if (name === 'a')
+                current = undefined;
+        },
+    }, { decodeEntities: true }).end(html);
+    const here = new URL(pageUrl);
+    let base = here.href;
+    try {
+        if (baseHref !== undefined)
+            base = new URL(baseHref, here).href;
+    }
+    catch { /* ignore invalid base */ }
+    return raw.flatMap(link => {
+        if (!link.href || link.href.startsWith('#'))
+            return [];
+        try {
+            const url = new URL(link.href, base);
+            if (['javascript:', 'data:', 'blob:'].includes(url.protocol))
+                return [];
+            if (sameOriginOnly && url.origin !== here.origin)
+                return [];
+            return [{ href: /^[a-z][a-z\d+.-]*:/i.test(link.href) ? link.href : url.href, text: link.text.trim().slice(0, 200) }];
+        }
+        catch {
+            return [];
+        }
+    });
+}
 /**
  * Collapse links that share an href (keeping the first, but preferring a
  * non-empty label) when `dedupe` is set, then cap to `limit`. Order is

package/dist/src/mcp/tools.js

@@ -857,7 +857,7 @@ exports.TOOL_HANDLERS = {
         const res = await ctx.ex.getText((0, validators_1.optionalTarget)(a), { tabId: tabId(a), ...frameOpts(a) });
         // Save the FULL read before capping — the artifact on disk stays lossless;
         // only what crosses into the caller's context is bounded.
-        (0, workspace_1.saveResult)('get_text', 'json', JSON.stringify(res, null, 2));
+        const savedTo = (0, workspace_1.saveResult)('get_text', 'json', JSON.stringify(res, null, 2));
         const clean = (0, redact_1.redactText)(res.text, redaction(ctx.policy));
         (0, audit_1.noteRedactions)(clean.redactions);
         const capped = (0, limits_1.capText)(clean.value, maxBytes(a));
@@ -865,6 +865,7 @@ exports.TOOL_HANDLERS = {
         return (0, envelopes_1.jsonResult)({
             ...res,
             text: capped.text,
+            ...(savedTo ? { savedTo } : {}),
             ...(0, limits_1.truncationMeta)(capped),
             ...(clean.redactions ? { redactions: clean.redactions } : {}),
         });
@@ -993,7 +994,7 @@ exports.TOOL_HANDLERS = {
             tabId: tabId(a),
             ...frameOpts(a),
         });
-        (0, workspace_1.saveResult)('read_as_markdown', 'md', md);
+        const savedTo = (0, workspace_1.saveResult)('read_as_markdown', 'md', md);
         const clean = (0, redact_1.redactText)(md, redaction(ctx.policy));
         (0, audit_1.noteRedactions)(clean.redactions);
         const capped = (0, limits_1.capText)(clean.value, maxBytes(a));
@@ -1001,7 +1002,7 @@ exports.TOOL_HANDLERS = {
         // Markdown is returned as plain text, so the truncation notice has to ride in
         // the body rather than as sibling JSON fields.
         return (0, envelopes_1.textResult)(capped.truncated
-            ? `${capped.text}\n\n[truncated: ${capped.returnedBytes} of ${capped.totalBytes} bytes — raise maxBytes or narrow with selector; the full document was saved to the task's results/ dir]`
+            ? `${capped.text}\n\n[truncated: ${capped.returnedBytes} of ${capped.totalBytes} bytes — raise maxBytes or narrow with selector${savedTo ? `; full document saved to ${savedTo}` : '; full document could not be saved'}]`
             : capped.text);
     },
     fill_form: async (a, ctx) => {

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "mcp-browser-extension",
   "mcpName": "io.github.Mehmoodqureshi/mcp-browser-extension",
-  "version": "0.9.26",
+  "version": "0.9.27",
   "description": "Drive your real Chrome browser over MCP \u2014 real logins, real cookies. A stdio MCP server (CLI) plus an MV3 extension, driving Chrome via chrome.scripting/chrome.tabs. Multi-tab batch automation, accessibility snapshots, deny-all security by default.",
   "author": "Mehmood Ur Rehman Qureshi",
   "license": "MIT",
@@ -61,7 +61,8 @@
     "prepack": "npm run clean && npm run build && npm run build:ext"
   },
   "dependencies": {
-    "@modelcontextprotocol/sdk": "^1.29.0",
+    "@modelcontextprotocol/sdk": "^1.32.1",
+    "htmlparser2": "^10.1.0",
     "ws": "^8.18.0",
     "zod": "^4.4.3"
   },