mail: PDF-Anhaenge als extrahierten Text zurueckgeben

read_attachment extrahiert PDF-Text serverseitig via pdfplumber und liefert
ihn als TextContent (statt EmbeddedResource-Blob, den claude.ai nicht lesen
kann). Bei Scan-/Bild-PDFs ohne Text: Hinweis + Blob-Fallback.
Loest 'Connector kann PDF nicht lesen' bei Buchungsbestaetigungen etc.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

mail/server.py

@@ -3,6 +3,7 @@
 import os
 import sys
 import base64
+import io
 import contextlib
 import imaplib
 import mailbox
@@ -252,7 +253,7 @@ def read_attachment(
     key: Annotated[str, Field(description="Message key from search results")],
     attachment_index: Annotated[int, Field(description="Attachment number from the read_mail attachment list (1-based)")],
 ) -> list[TextContent | ImageContent | EmbeddedResource]:
-    """Read an email attachment. Images shown inline, documents (PDF/docx) as binary, text directly. Get the index from read_mail."""
+    """Read an email attachment. Images shown inline, PDFs as extracted text, text directly, other documents as binary. Get the index from read_mail."""
     user = get_current_user()
     if not user:
         return [TextContent(type="text", text="Error: not authenticated")]
@@ -275,6 +276,27 @@ def read_attachment(
         return [ImageContent(type="image", data=base64.b64encode(payload).decode(), mimeType=mime)]
     if mime.startswith("text/"):
         return [TextContent(type="text", text=payload.decode("utf-8", errors="replace")[:100000])]
+    if mime == "application/pdf" or att["filename"].lower().endswith(".pdf"):
+        try:
+            import pdfplumber
+            with pdfplumber.open(io.BytesIO(payload)) as pdf:
+                pages = []
+                for i, page in enumerate(pdf.pages, 1):
+                    t = page.extract_text() or ""
+                    if t.strip():
+                        pages.append(f"--- Seite {i} ---\n{t}")
+                text = "\n\n".join(pages)
+            if text.strip():
+                return [TextContent(type="text", text=f"[PDF: {att['filename']}]\n\n{text[:200000]}")]
+            # Kein extrahierbarer Text -> vermutlich Scan/Bild-PDF: als Blob zurueck
+            return [
+                TextContent(type="text", text=f"[PDF '{att['filename']}' enthaelt keinen extrahierbaren Text (vermutlich Scan). Rohdaten folgen.]"),
+                EmbeddedResource(type="resource", resource=BlobResourceContents(
+                    uri=f"mail://attachment/{att['filename']}",
+                    blob=base64.b64encode(payload).decode(), mimeType=mime)),
+            ]
+        except Exception as e:
+            return [TextContent(type="text", text=f"PDF '{att['filename']}' konnte nicht gelesen werden: {e}")]
     return [EmbeddedResource(type="resource", resource=BlobResourceContents(
         uri=f"mail://attachment/{att['filename']}",
         blob=base64.b64encode(payload).decode(), mimeType=mime))]