From b88adc4c50d435e660ed3053aca406af334d0fe9 Mon Sep 17 00:00:00 2001
From: Stefan Lohmaier <stefan@slohmaier.de>
Date: Fri, 12 Jun 2026 08:02:32 +0200
Subject: [PATCH] read_file: pass binary documents as EmbeddedResource
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

PDF, docx, xlsx, pptx etc. are returned as EmbeddedResource with
BlobResourceContents (base64 + mimeType). claude.ai processes these
client-side, same as Google Drive/Dropbox MCP connectors.

Fallback chain: text → image → embedded resource (binary).
MIME type guessed from extension when server returns octet-stream.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 files/server.py | 44 ++++++++++++++++++++++++++++++++++++--------
 1 file changed, 36 insertions(+), 8 deletions(-)

diff --git a/files/server.py b/files/server.py
index 6382f2f..ceb2b3b 100644
--- a/files/server.py
+++ b/files/server.py
@@ -7,7 +7,7 @@ from typing import Annotated
 import httpx
 from pydantic import Field
 from mcp.server.fastmcp import FastMCP
-from mcp.types import TextContent, ImageContent
+from mcp.types import TextContent, ImageContent, EmbeddedResource, BlobResourceContents
 from starlette.applications import Starlette
 from starlette.routing import Mount
 
@@ -71,25 +71,53 @@ def list_files(
 
 IMAGE_TYPES = {"image/jpeg", "image/png", "image/gif", "image/webp", "image/svg+xml"}
 TEXT_HINTS = ["text/", "json", "xml", "csv", "yaml", "javascript", "markdown"]
+DOC_TYPES = {
+    "application/pdf", "application/msword",
+    "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
+    "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
+    "application/vnd.openxmlformats-officedocument.presentationml.presentation",
+    "application/vnd.ms-excel", "application/vnd.ms-powerpoint",
+    "application/rtf", "application/epub+zip",
+}
+MAX_BIN_SIZE = 25_000_000
+
+def _guess_mime(path, ct):
+    if ct and ct != "application/octet-stream":
+        return ct
+    ext = path.rsplit(".", 1)[-1].lower() if "." in path else ""
+    return {
+        "pdf": "application/pdf", "docx": "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
+        "xlsx": "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
+        "pptx": "application/vnd.openxmlformats-officedocument.presentationml.presentation",
+        "doc": "application/msword", "xls": "application/vnd.ms-excel", "ppt": "application/vnd.ms-powerpoint",
+        "rtf": "application/rtf", "jpg": "image/jpeg", "jpeg": "image/jpeg", "png": "image/png",
+        "gif": "image/gif", "webp": "image/webp", "svg": "image/svg+xml",
+    }.get(ext, ct or "application/octet-stream")
 
 @mcp.tool()
 def read_file(
-    path: Annotated[str, Field(description="Full file path, e.g. '/Documents/notes.txt', '/Photos/pic.jpg'")],
-) -> list[TextContent | ImageContent]:
-    """Read a file. Text files return content directly. Images (jpg/png/gif/webp) are displayed inline. Other binary files return only metadata."""
+    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 are displayed inline. Documents (PDF, docx, xlsx, pptx) are passed as binary for the client to process. 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)
     if r.status_code >= 400: return [TextContent(type="text", text=f"Fehler: HTTP {r.status_code}")]
-    ct = r.headers.get("content-type", "").split(";")[0].strip()
-    if ct in IMAGE_TYPES and len(r.content) < 10_000_000:
+    ct = _guess_mime(path, r.headers.get("content-type", "").split(";")[0].strip())
+    size = len(r.content)
+    if size > MAX_BIN_SIZE:
+        return [TextContent(type="text", text=f"Datei zu gross: {size:,} bytes (max {MAX_BIN_SIZE:,}). Typ: {ct}")]
+    if ct in IMAGE_TYPES:
         return [ImageContent(type="image", data=base64.b64encode(r.content).decode(), mimeType=ct)]
     if any(t in ct for t in TEXT_HINTS):
         return [TextContent(type="text", text=r.text[:100000])]
     try:
-        return [TextContent(type="text", text=r.content.decode("utf-8")[:100000])]
+        text = r.content.decode("utf-8")
+        return [TextContent(type="text", text=text[:100000])]
     except Exception:
-        return [TextContent(type="text", text=f"Binaerdatei ({len(r.content):,} bytes, Typ: {ct}). Nutze read_file nur fuer Text und Bilder.")]
+        pass
+    b64 = base64.b64encode(r.content).decode()
+    return [EmbeddedResource(type="resource", resource=BlobResourceContents(uri=f"file://{path}", blob=b64, mimeType=ct))]
 
 
 @mcp.tool()