diff --git a/src/late/mcp/chatgpt_server.py b/src/late/mcp/chatgpt_server.py index db0cb746..95584cb2 100644 --- a/src/late/mcp/chatgpt_server.py +++ b/src/late/mcp/chatgpt_server.py @@ -25,6 +25,7 @@ from late.client.base import _retry_after_seconds from late.mcp.auth import build_auth_provider, forget_rejected_token +from late.mcp.media_items import media_item from late.mcp.middleware import WellFormedOutputMiddleware, WellFormedOutputTransform from late.mcp.middleware.rate_limit import wait_hint from late.mcp.user_agent import mcp_user_agent @@ -47,7 +48,6 @@ "Saturday", "Sunday", ] -_VIDEO_EXT = (".mp4", ".mov", ".avi", ".webm", ".m4v") # Everything this surface can do fits these four; ChatGPT asks the user to # grant every scope the discovery document advertises. CHATGPT_SCOPES = ["posts:read", "posts:write", "accounts:read", "analytics:read"] @@ -449,19 +449,7 @@ async def _usernames(posts: list[dict[str, Any]]) -> dict[str, str]: def _media_items(media_urls: list[str] | None) -> list[dict[str, str]]: - items = [] - for url in media_urls or []: - lower = url.lower().split("?")[0] - if lower.endswith(_VIDEO_EXT): - media_type = "video" - elif lower.endswith(".gif"): - media_type = "gif" - elif lower.endswith(".pdf"): - media_type = "document" - else: - media_type = "image" - items.append({"type": media_type, "url": url}) - return items + return [media_item(url) for url in media_urls or []] def _schedule_fields(scheduled_for: str, tz: str | None) -> dict[str, str]: diff --git a/src/late/mcp/media_items.py b/src/late/mcp/media_items.py new file mode 100644 index 00000000..65c52e28 --- /dev/null +++ b/src/late/mcp/media_items.py @@ -0,0 +1,18 @@ +from __future__ import annotations + +_VIDEO_EXT = (".mp4", ".mov", ".avi", ".webm", ".m4v") +_IMAGE_EXT = (".jpg", ".jpeg", ".png", ".webp", ".heic", ".heif", ".bmp", ".tiff") + + +def media_item(url: str) -> dict[str, str]: + path = url.lower().split("?")[0] + if path.endswith(_VIDEO_EXT): + return {"type": "video", "url": url} + if path.endswith(".gif"): + return {"type": "gif", "url": url} + if path.endswith(".pdf"): + return {"type": "document", "url": url} + if path.endswith(_IMAGE_EXT): + return {"type": "image", "url": url} + # Unknown extension (e.g. a Drive link): omit type so the API infers it (Drive links are rehosted first). + return {"url": url} diff --git a/src/late/mcp/server.py b/src/late/mcp/server.py index d33cde39..b73fc37f 100644 --- a/src/late/mcp/server.py +++ b/src/late/mcp/server.py @@ -45,11 +45,12 @@ from fastmcp.server.transforms.search.base import serialize_tools_for_output_markdown from mcp.types import INVALID_PARAMS, ToolAnnotations -from late import Late, MediaType, PostStatus +from late import Late, PostStatus from late.client.exceptions import LateAPIError, LateError, LateRateLimitError from .auth import build_auth_provider, forget_rejected_token from .idempotency import same_arguments_key +from .media_items import media_item from .middleware import ( FEEDBACK_TOOL, FeedbackSourceMiddleware, @@ -931,16 +932,7 @@ def posts_create( # Add media items if provided if media_urls: urls = [u.strip() for u in media_urls.split(",") if u.strip()] - media_items = [] - for url in urls: - media_type: MediaType | str = MediaType.IMAGE - if any( - ext in url.lower() for ext in [".mp4", ".mov", ".avi", ".webm", ".m4v"] - ): - media_type = MediaType.VIDEO - elif any(ext in url.lower() for ext in [".gif"]): - media_type = MediaType.GIF - media_items.append({"type": media_type, "url": url}) + media_items = [media_item(url) for url in urls] params["media_items"] = media_items if is_draft: @@ -1042,16 +1034,7 @@ def posts_cross_post( if media_urls: urls = [u.strip() for u in media_urls.split(",") if u.strip()] - media_items = [] - for url in urls: - media_type: MediaType | str = MediaType.IMAGE - if any( - ext in url.lower() for ext in [".mp4", ".mov", ".avi", ".webm", ".m4v"] - ): - media_type = MediaType.VIDEO - elif any(ext in url.lower() for ext in [".gif"]): - media_type = MediaType.GIF - media_items.append({"type": media_type, "url": url}) + media_items = [media_item(url) for url in urls] params["media_items"] = media_items if is_draft: diff --git a/tests/test_chatgpt_server.py b/tests/test_chatgpt_server.py index 5246d14f..91491492 100644 --- a/tests/test_chatgpt_server.py +++ b/tests/test_chatgpt_server.py @@ -18,7 +18,7 @@ from fastmcp import Client from late.mcp import auth, chatgpt_server -from late.mcp.chatgpt_server import chatgpt_mcp +from late.mcp.chatgpt_server import _media_items, chatgpt_mcp API = chatgpt_server.API_BASE @@ -341,3 +341,22 @@ async def test_an_api_401_evicts_the_cached_token_so_chatgpt_gets_a_401_and_refr assert result.is_error assert auth.token_cache_key("sk_test") not in auth._VERIFIED_AT + + +@pytest.mark.parametrize( + ("url", "expected"), + [ + ("https://x.com/a.mp4", {"type": "video", "url": "https://x.com/a.mp4"}), + ( + "https://x.com/a.jpg?w=1", + {"type": "image", "url": "https://x.com/a.jpg?w=1"}, + ), + ("https://x.com/a.pdf", {"type": "document", "url": "https://x.com/a.pdf"}), + ( + "https://drive.google.com/file/d/abc/view?usp=sharing", + {"url": "https://drive.google.com/file/d/abc/view?usp=sharing"}, + ), + ], +) +def test_media_items_only_type_what_the_extension_proves(url, expected): + assert _media_items([url]) == [expected] diff --git a/tests/test_mcp_media_items.py b/tests/test_mcp_media_items.py new file mode 100644 index 00000000..fcd4b654 --- /dev/null +++ b/tests/test_mcp_media_items.py @@ -0,0 +1,59 @@ +from __future__ import annotations + +import json + +import httpx +import pytest +import respx +from fastmcp import Client + +from late.mcp.server import mcp + +API = "https://api.test.com" +DRIVE_URL = "https://drive.google.com/file/d/abc/view?usp=sharing" +POST_RESPONSE = { + "post": {"_id": "post_1", "content": "hi", "status": "draft", "platforms": []}, + "message": "ok", +} + + +@pytest.fixture(autouse=True) +def _api_env(monkeypatch: pytest.MonkeyPatch) -> None: + monkeypatch.setenv("ZERNIO_API_KEY", "test_api_key") + monkeypatch.setenv("ZERNIO_BASE_URL", API) + + +@respx.mock +@pytest.mark.parametrize( + ("tool", "arguments"), + [ + ( + "posts_create", + {"platform": "twitter", "account_id": "acc_1"}, + ), + ( + "posts_cross_post", + {"platforms": "twitter", "account_ids": "acc_1"}, + ), + ], +) +async def test_media_urls_are_typed_only_when_the_extension_proves_it( + tool: str, arguments: dict +) -> None: + create = respx.post(f"{API}/v1/posts").mock( + return_value=httpx.Response(200, json=POST_RESPONSE) + ) + media_urls = f"https://x.com/a.mp4,https://x.com/a.jpg,{DRIVE_URL}" + + async with Client(mcp) as client: + await client.call_tool( + tool, + {"content": "hi", "is_draft": True, "media_urls": media_urls, **arguments}, + ) + + items = json.loads(create.calls.last.request.content)["mediaItems"] + assert items == [ + {"type": "video", "url": "https://x.com/a.mp4"}, + {"type": "image", "url": "https://x.com/a.jpg"}, + {"url": DRIVE_URL}, + ]