fix: restore PyPDF2 as the PDF text extractor in the SDK path

Text extraction (and therefore tree building) now matches the CLI /
pre-SDK default. Image extraction keeps PyMuPDF — the only part that
needs it — with ![image](path) references appended per page.

README.md

@@ -205,7 +205,7 @@ Omitting `doc_ids` queries the **entire collection** and lets the agent pick whi
 
 # ⚙️ Package Usage
 
-> **Note:** This package uses standard PDF parsing. The CLI parses PDFs with PyPDF2 (as in previous releases), while the SDK's local mode parses with PyMuPDF and also extracts images — extracted text can differ slightly, so the two entry points may produce different trees for the same PDF. For use cases with complex PDFs, our [cloud service](https://pageindex.ai/developer) (via MCP and API) offers enhanced OCR, tree building, and retrieval.
+> **Note:** This package uses standard PDF parsing. For use cases with complex PDFs, our [cloud service](https://pageindex.ai/developer) (via MCP and API) offers enhanced OCR, tree building, and retrieval.
 
 You can follow these steps to generate a PageIndex tree from a PDF document.
 

pageindex/parser/pdf.py

@@ -1,3 +1,4 @@
+import PyPDF2
 import pymupdf
 from pathlib import Path
 from .protocol import ContentNode, ParsedDocument
@@ -15,39 +16,42 @@ class PdfParser:
         path = Path(file_path)
         model = kwargs.get("model")
         images_dir = kwargs.get("images_dir")
+
+        # Images are extracted with PyMuPDF (PyPDF2 cannot); text stays with
+        # PyPDF2 below so the extracted text — and therefore the tree — matches
+        # the CLI / pre-SDK default.
+        page_images: dict[int, list[dict]] = {}
+        if images_dir:
+            with pymupdf.open(str(path)) as doc:
+                for i, page in enumerate(doc):
+                    images = self._extract_page_images(page, i + 1, images_dir)
+                    if images:
+                        page_images[i + 1] = images
+
+        reader = PyPDF2.PdfReader(str(path))
         nodes = []
+        for i, page in enumerate(reader.pages):
+            page_num = i + 1
+            content = page.extract_text() or ""
+            images = page_images.get(page_num)
+            if images:
+                refs = "\n".join(f"![image]({img['path']})" for img in images)
+                content = f"{content}\n{refs}" if content else refs
 
-        with pymupdf.open(str(path)) as doc:
+            tokens = count_tokens(content, model=model)
-            page_count = doc.page_count
+            nodes.append(ContentNode(
-            for i, page in enumerate(doc):
+                content=content,
-                page_num = i + 1
+                tokens=tokens,
-                if images_dir:
+                index=page_num,
-                    content, images = self._extract_page_with_images(
+                images=images,
-                        doc, page, page_num, images_dir)
+            ))
-                else:
-                    content = page.get_text()
-                    images = None
-
-                tokens = count_tokens(content, model=model)
-                nodes.append(ContentNode(
-                    content=content or "",
-                    tokens=tokens,
-                    index=page_num,
-                    images=images if images else None,
-                ))
 
         return ParsedDocument(doc_name=path.stem, nodes=nodes,
-                              metadata={"page_count": page_count})
+                              metadata={"page_count": len(reader.pages)})
 
     @staticmethod
-    def _extract_page_with_images(doc, page, page_num: int,
+    def _extract_page_images(page, page_num: int, images_dir: str) -> list[dict]:
-                                  images_dir: str) -> tuple[str, list[dict]]:
+        """Save a page's images to disk and return their metadata."""
-        """Extract text and images from a page, preserving their relative order.
-
-        Uses get_text("dict") to iterate blocks in reading order.
-        Text blocks become text; image blocks are saved to disk and replaced
-        with an inline placeholder: ![image](path)
-        """
         images_path = Path(images_dir)
         images_path.mkdir(parents=True, exist_ok=True)
         # Store an absolute path so the ![image](...) reference resolves
@@ -55,53 +59,42 @@ class PdfParser:
         # break as soon as the query runs from a different directory.)
         abs_images_path = images_path.resolve()
 
-        parts: list[str] = []
         images: list[dict] = []
         img_idx = 0
 
         for block in page.get_text("dict")["blocks"]:
-            if block["type"] == 0:  # text block
+            if block["type"] != 1:  # image blocks only
-                lines = []
+                continue
-                for line in block["lines"]:
+            width = block.get("width", 0)
-                    spans_text = "".join(span["text"] for span in line["spans"])
+            height = block.get("height", 0)
-                    lines.append(spans_text)
+            if width < _MIN_IMAGE_SIZE or height < _MIN_IMAGE_SIZE:
-                parts.append("\n".join(lines))
+                continue
 
-            elif block["type"] == 1:  # image block
+            image_bytes = block.get("image")
-                width = block.get("width", 0)
+            if not image_bytes:
-                height = block.get("height", 0)
+                continue
-                if width < _MIN_IMAGE_SIZE or height < _MIN_IMAGE_SIZE:
-                    continue
 
-                image_bytes = block.get("image")
+            try:
-                if not image_bytes:
+                pix = pymupdf.Pixmap(image_bytes)
-                    continue
+                # n includes the alpha channel, so a plain RGBA pixmap also
+                # has n==4 — subtract alpha before comparing. Without this,
+                # a CMYK image with no alpha (n==4, same as RGBA) skips the
+                # RGB conversion, and pix.save() as .png then raises
+                # "unsupported colorspace for 'png'", silently dropping the
+                # image via the bare except below.
+                if pix.n - pix.alpha >= 4:
+                    pix = pymupdf.Pixmap(pymupdf.csRGB, pix)
+                filename = f"p{page_num}_img{img_idx}.png"
+                pix.save(str(images_path / filename))
+                pix = None
+            except Exception:
+                continue
 
-                try:
+            images.append({
-                    pix = pymupdf.Pixmap(image_bytes)
+                "path": str(abs_images_path / filename),
-                    # n includes the alpha channel, so a plain RGBA pixmap also
+                "width": width,
-                    # has n==4 — subtract alpha before comparing. Without this,
+                "height": height,
-                    # a CMYK image with no alpha (n==4, same as RGBA) skips the
+            })
-                    # RGB conversion, and pix.save() as .png then raises
+            img_idx += 1
-                    # "unsupported colorspace for 'png'", silently dropping the
-                    # image via the bare except below.
-                    if pix.n - pix.alpha >= 4:
-                        pix = pymupdf.Pixmap(pymupdf.csRGB, pix)
-                    filename = f"p{page_num}_img{img_idx}.png"
-                    save_path = images_path / filename
-                    pix.save(str(save_path))
-                    pix = None
-                except Exception:
-                    continue
 
-                img_path = str(abs_images_path / filename)
+        return images
-                images.append({
-                    "path": img_path,
-                    "width": width,
-                    "height": height,
-                })
-                parts.append(f"![image]({img_path})")
-                img_idx += 1
-
-        content = "\n".join(parts)
-        return content, images