feat: Office-Extraktion (docx/xlsx/pptx) fuer files-, mail- und notes-MCP

officeutil.py (analog pdfutil): server-seitige Text-Extraktion, damit das
LLM Inhalte als Text bekommt statt Base64-Blob. Verkabelt in files.read_file,
mail.read_attachment (und notes.read_resource, bereits in e34711d).
Aeltere Binaerformate (.doc/.xls/.ppt) und .rtf bewusst nicht behandelt --
is_office() False -> Fallback auf Rohdaten. requirements-extra: python-pptx.

War bereits deployed und lief (FileTypes-Tests 14 passed), lag aber
uncommittet im Arbeitsverzeichnis; ohne officeutil.py waere ein frischer
Clone seit e34711d kaputt gewesen.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XhHNJKk3RsHgUsJ27WQSom

files/server.py

@@ -15,6 +15,7 @@ from starlette.routing import Mount
 sys.path.insert(0, os.path.dirname(os.path.dirname(__file__)))
 from common import get_current_user, OAUTH_ROUTES, BearerAuthMiddleware
 from pdfutil import pdf_to_content
+from officeutil import office_to_content, is_office
 
 from common import load_config as _lc
 _cfg = _lc()
@@ -104,7 +105,7 @@ def _guess_mime(path, ct):
 def read_file(
     path: Annotated[str, Field(description="Full file path, e.g. '/Documents/notes.txt', '/report.pdf', '/photo.jpg'")],
 ) -> list[TextContent | ImageContent | EmbeddedResource]:
-    """Read a file. Text files return content directly. Images inline. PDFs as extracted text. Other documents (docx, xlsx, pptx) as binary. Max 25 MB."""
+    """Read a file. Text files return content directly. Images inline. PDFs and Office documents (docx, xlsx, pptx) as extracted text. Max 25 MB."""
     user = get_current_user()
     if not user: return [TextContent(type="text", text="Error: not authenticated")]
     r = httpx.get(_dav(user, path), auth=_auth(user), timeout=60)
@@ -119,6 +120,8 @@ def read_file(
         return [TextContent(type="text", text=r.text[:100000])]
     if ct == "application/pdf" or path.lower().endswith(".pdf"):
         return pdf_to_content(r.content, path, ct)
+    if is_office(ct, path):
+        return office_to_content(r.content, path, ct)
     try:
         text = r.content.decode("utf-8")
         return [TextContent(type="text", text=text[:100000])]