@page2ai/mcp 0.4.0 → 0.4.1

package/CHANGELOG.md

@@ -5,6 +5,24 @@ The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/) and
 project uses [Semantic Versioning](https://semver.org/spec/v2.0.0.html); while the
 major version is `0`, a minor bump may carry breaking changes.
 
+## 0.4.1 - 2026-10-20
+
+### Fixed
+- **`@page2ai/core` 0.3.1: layout tables convert as blocks, not as one table cell** (P2A-024), and
+  the dependency range starts there (`^0.3.1`, was `^0.3.0`). An HTML e-mail or newsletter built
+  from nested `<table role="presentation">` came out as one Markdown table cell on one line, its
+  heading and paragraphs run together (the IronMemo letter, one 1,415-character cell holding its
+  `<h1>` and 14 paragraphs). Such a table now converts as the blocks in its cells, in source order;
+  a data table inside it stays a table. Images in a layout table keep the cell rule, so spacers and
+  open-tracking pixels (empty alt) add nothing. The range moves for the reason given in 0.3.1: with
+  a warm npx cache `page2ai-mcp` 0.4.0 still ran core 0.3.0.
+- **`@page2ai/mcp` 0.4.0, P2A-023: the hosted endpoint clamps `timeout_ms`** to 20 s under the
+  30 s Vercel function limit. A request with `timeout_ms` above 20 000 previously kept the
+  connection alive past the function's hard limit and returned only keepalive comments.
+- **SECURITY.md** no longer says the DNS-rebinding window is open (P2A-022). Core 0.3.0 closed it
+  with DNS pinning; the published SECURITY.md of 0.4.0 still described it as a known limitation.
+  Supported-versions table updated to 0.4.x.
+
 ## 0.4.0 - 2026-10-18
 
 ### Security

package/manifest.json

@@ -2,7 +2,7 @@
   "manifest_version": "0.3",
   "name": "page2ai-mcp",
   "display_name": "Page2AI — Web to Markdown",
-  "version": "0.4.0",
+  "version": "0.4.1",
   "description": "Convert any web page to clean Markdown for LLM context",
   "long_description": "Page2AI MCP converts a web page URL to Markdown for AI context windows. It keeps the page's main content (its <main> or dominant <article>), preserves code blocks with language hints, converts tables and lists, and drops navigation, headers, footers and sidebars. The only network requests go to the URL you pass it and the addresses it redirects to, asked for as a browser opens a page, so docs platforms return the page itself rather than a Markdown copy of it: pure Node + linkedom, no browser, no cookies, no page scripts, no external services, no telemetry. A page behind a sign-in, or one rendered only by JavaScript, is therefore out of its reach.\n\nCompanion to the Page2AI browser extension (https://github.com/igorsaevets/page2ai-extension), which converts the live page in your own browser, including pages behind a sign-in. Built on the open-source @page2ai/core library.",
   "author": {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@page2ai/mcp",
-  "version": "0.4.0",
+  "version": "0.4.1",
   "mcpName": "io.github.igorsaevets/page2ai-mcp",
   "description": "MCP server that converts any web page URL to clean Markdown for LLM context. Processing is local and SSRF-guarded; no external APIs, no telemetry.",
   "type": "module",
@@ -35,7 +35,7 @@
   },
   "dependencies": {
     "@modelcontextprotocol/server": "^2.0.0",
-    "@page2ai/core": "^0.3.0",
+    "@page2ai/core": "^0.3.1",
     "zod": "^4.2.0"
   },
   "devDependencies": {

package/server.json

@@ -8,12 +8,12 @@
     "source": "github"
   },
   "websiteUrl": "https://github.com/igorsaevets/page2ai-mcp#readme",
-  "version": "0.4.0",
+  "version": "0.4.1",
   "packages": [
     {
       "registryType": "npm",
       "identifier": "@page2ai/mcp",
-      "version": "0.4.0",
+      "version": "0.4.1",
       "transport": {
         "type": "stdio"
       }