Files in ChatGPT Apps (MCP): parity with GPT file Actions?

Hi all,

I’m building a ChatGPT App using the new MCP Apps framework and I’m trying to support this flow:

  1. User pastes or uploads an image directly in ChatGPT.

  2. The model interprets the image (vision), extracts structured data.

  3. My app’s tool is called with both:

    • the extracted structured data (JSON), and

    • a handle to the original media file (e.g., URL or file ID) so my backend can store/process it.

In Custom GPTs, this is basically what GPT file Actions provide: the model fills JSON arguments and the Action receives attached files via short‑lived URLs in a single call.

Questions:

  • Is there an equivalent or planned feature for Apps/MCP tools, where an MCP tool can receive both structured arguments and user‑uploaded files from the chat context in one invocation?

  • If not today, is there any documented roadmap or recommended pattern (beyond building my own upload UI in a widget) to get parity with GPT file Actions for Apps?

Thanks!

I’ve been digging into MCP pretty heavily lately, and from what I’m seeing, tools still don’t receive file attachments the way GPT Actions do. The model can parse the image and hand me structured JSON just fine, but the raw file never reaches my tool call.

File Actions are not currently supported. In our memorial video editor, we solved this by instructing the user to upload the file in the widget, not in the conversation.

This way, we can render the video slideshow at the original file quality (see UX demo here).

But I agree, it would definitely be easier if GPT File Actions were available.

1 Like