feat(mcp): download any message file via share link

backend/src/api/mcp/download.py

@@ -0,0 +1,113 @@
+import asyncio
+from pathlib import Path
+from typing import Any
+
+import asyncpg
+from fastmcp.exceptions import ToolError
+from fastmcp.server.dependencies import get_http_request
+from fastmcp.utilities.types import Image
+from mcp.types import TextContent
+
+from utils.files import expiry_from_seconds
+from utils.jobs import enqueue
+from utils.read import chats, media, shares
+from utils.read.models import FileShareView, MediaView
+from utils.storage import ContentAddressedStorage
+
+LINK_TTL_SECONDS = 24 * 3600
+MAX_WAIT_SECONDS = 120
+_POLL_SECONDS = 1.0
+_INLINE_IMAGE_BYTES = 5 * 1024 * 1024
+_INLINE_TEXT_BYTES = 256 * 1024
+_INLINE_IMAGE_MIMES = {"image/jpeg", "image/png", "image/gif", "image/webp"}
+_TEXT_MIMES = {"application/json", "application/xml", "application/x-yaml"}
+_FAILED_JOB = ("failed", "canceled")
+
+
+async def _job_error(pool: asyncpg.Pool, job_id: int) -> str | None:
+    row = await pool.fetchrow("SELECT status, error FROM jobs WHERE id = $1", job_id)
+    if row is None or row["status"] in _FAILED_JOB:
+        return (row and row["error"]) or "download job is gone"
+    return None
+
+
+async def fetch_media(
+    pool: asyncpg.Pool,
+    account_id: int,
+    chat_id: int,
+    message_id: int,
+    wait_seconds: float,
+) -> MediaView | None:
+    item = await media.get_message_media(pool, account_id, chat_id, message_id)
+    if item is None:
+        msg = f"message {chat_id}/{message_id} has no media"
+        raise ToolError(msg)
+    if item.downloaded and item.storage_key:
+        return item
+    job_id = await enqueue(
+        pool, account_id, "fetch_media", {"chat_id": chat_id, "message_id": message_id}
+    )
+    loop = asyncio.get_running_loop()
+    deadline = loop.time() + wait_seconds
+    while loop.time() < deadline:
+        await asyncio.sleep(_POLL_SECONDS)
+        item = await media.get_message_media(pool, account_id, chat_id, message_id)
+        if item is not None and item.downloaded and item.storage_key:
+            return item
+        error = await _job_error(pool, job_id)
+        if error is not None:
+            msg = f"download failed: {error}"
+            raise ToolError(msg)
+    return None
+
+
+def _origin() -> str:
+    request = get_http_request()
+    proto = request.headers.get("x-forwarded-proto", request.url.scheme)
+    return f"{proto}://{request.url.netloc}"
+
+
+async def share_media(pool: asyncpg.Pool, item: MediaView, mime: str) -> FileShareView:
+    storage_key = item.storage_key or ""
+    share = await shares.find_active_share(pool, item.account_id, storage_key)
+    if share is None:
+        chat = await chats.get_chat(pool, item.account_id, item.chat_id)
+        share = await shares.create_share(
+            pool,
+            item.account_id,
+            "media",
+            storage_key,
+            item.file_name or f"media_{item.id}",
+            mime=mime,
+            file_size=item.file_size,
+            title=chat.title if chat else None,
+            chat_id=item.chat_id,
+            message_id=item.message_id,
+            expires_at=expiry_from_seconds(LINK_TTL_SECONDS),
+        )
+    return share
+
+
+def _is_text(mime: str) -> bool:
+    return mime.startswith("text/") or mime in _TEXT_MIMES
+
+
+def inline_content(storage: ContentAddressedStorage, key: str, mime: str) -> list[Any]:
+    size = Path(storage.url(key)).stat().st_size
+    if mime in _INLINE_IMAGE_MIMES and size <= _INLINE_IMAGE_BYTES:
+        return [Image(data=storage.get(key), format=mime.rsplit("/", maxsplit=1)[-1])]
+    if _is_text(mime) and size <= _INLINE_TEXT_BYTES:
+        text = storage.get(key).decode("utf-8", errors="replace")
+        return [TextContent(type="text", text=text)]
+    return []
+
+
+def describe(item: MediaView, mime: str, share: FileShareView) -> str:
+    size = f"{item.file_size} bytes" if item.file_size else "size unknown"
+    expires = share.expires_at.isoformat() if share.expires_at else "never"
+    return (
+        f"{item.file_name} ({item.kind}, {mime}, {size})\n"
+        f"Download: {_origin()}/f/{share.token}/{share.url_name}\n"
+        f"Public link, expires {expires}; fetch it with curl/wget or any HTTP "
+        "client, no auth needed."
+    )

backend/src/api/mcp/server.py

@@ -7,12 +7,19 @@ from fastmcp import FastMCP
 from fastmcp.exceptions import ToolError
 from fastmcp.server.middleware import CallNext, Middleware, MiddlewareContext
 from fastmcp.tools.base import ToolResult
-from mcp.types import CallToolRequestParams
+from mcp.types import CallToolRequestParams, TextContent
 from pydantic import BaseModel
 
+from api.mcp.download import (
+    MAX_WAIT_SECONDS,
+    describe,
+    fetch_media,
+    inline_content,
+    share_media,
+)
 from api.mcp.format import build_transcript, load_notes, load_photos, resolve_names
 from dependencies.container import container
-from utils.jobs import enqueue
+from utils.files import resolve_mime
 from utils.read import annotations, chats, media, peers, presence, social, watches
 from utils.read.accounts import self_user_id, unknown_account_hint
 from utils.read.models import DEFAULT_LIMIT, Page
@@ -44,7 +51,7 @@ Typical flows:
   `before_id` (first id you saw), to newer with `after_id` (last id you saw);
   begin a full forward walk with `after_id=0`. Each transcript line starts
   with its `#message_id`, and photos show as `[photo #message_id]` — pass that
-  id to `get_media` (with `fetch=True` to download) or to annotation tools.
+  id to `download_media` or to annotation tools.
   Lines marked
   "📝 [your private note, NOT in Telegram]" are the user's own annotations
   attached locally via the web UI; they never existed in Telegram. Treat them
@@ -53,8 +60,11 @@ Typical flows:
   transcripts; supports chat/sender/date filters and regex).
 - Forensics: `get_deleted_messages` and `get_message_versions` recover content
   removed or edited in Telegram but kept in the archive.
-- Media: `get_media` returns metadata; pass `fetch=True` to enqueue a lazy
+- Files: `download_media` gets any message attachment (photo, video, voice,
-  download if the file isn't stored yet.
+  document, ...). It pulls the file from Telegram if the archive lacks it and
+  returns a public link valid 24h that needs no auth (`curl -O <link>`), plus
+  the content itself for small images and text files. `get_media` returns
+  metadata only.
 - Monitoring: `set_watch` creates a local rule, `list_watches` lists rules, and
   `list_alerts` reads what those rules fired.
 
@@ -63,7 +73,8 @@ Watched bots:
   own `account_id` and its chats, messages and media read through the same
   tools as a user account. Nothing bot-specific to call.
 
-Writes: everything is read-only except `set_watch`, the only allowed write.
+Writes: everything is read-only except `set_watch` and the share links
+`download_media` creates (visible and revocable in the web UI).
 
 Paging: list/search tools accept `limit` and `offset`.
 """
@@ -239,21 +250,40 @@ async def get_message_versions(
 
 @mcp.tool
 async def get_media(
-    account_id: int, chat_id: int, message_id: int, fetch: bool = False
+    account_id: int, chat_id: int, message_id: int
 ) -> dict[str, Any] | None:
-    """Get media metadata for a message; set fetch=True to enqueue lazy download."""
+    """Get media metadata for a message; use download_media to get the file."""
+    item = await media.get_message_media(await _pool(), account_id, chat_id, message_id)
+    return item.model_dump(mode="json") if item else None
+
+
+@mcp.tool(output_schema=None)
+async def download_media(
+    account_id: int, chat_id: int, message_id: int, wait_seconds: int = 45
+) -> list[Any]:
+    """Download the file attached to a message: any photo, video, voice, document.
+
+    Pulls it from Telegram first if the archive doesn't have it yet, waiting up
+    to `wait_seconds`. Returns a public download link (valid 24h, no auth) and
+    inlines small images and text files.
+    """
     pool = await _pool()
-    item = await media.get_message_media(pool, account_id, chat_id, message_id)
+    wait = min(max(wait_seconds, 0), MAX_WAIT_SECONDS)
-    if item is None:
+    item = await fetch_media(pool, account_id, chat_id, message_id, wait)
-        return None
+    if item is None or item.storage_key is None:
-    if fetch and not item.downloaded:
+        return [
-        await enqueue(
+            TextContent(
-            pool,
+                type="text",
-            account_id,
+                text="Still downloading from Telegram; call download_media again.",
-            "fetch_media",
+            )
-            {"chat_id": chat_id, "message_id": message_id},
+        ]
-        )
+    mime = resolve_mime(item.kind, item.mime, item.file_name)
-    return item.model_dump(mode="json")
+    share = await share_media(pool, item, mime)
+    storage = await container.get(ContentAddressedStorage)
+    return [
+        TextContent(type="text", text=describe(item, mime, share)),
+        *inline_content(storage, item.storage_key, mime),
+    ]
 
 
 @mcp.tool