How to handle files

[DamyanBG]

Question

How I can handle files with Agent or Agents? For example let's say I am sending an image for color recognition and then the Agent have to use MCP tool call to add this image to the relevant color. Unfortunately, my implementation forces the LLM to provide to the MCP tool call the URL link of the image, which is on OpenAI storage.
I think it would me good to support flows especially with images, where for example the Agent can see what is in the image , make changes to the image and then call tools.

[seratch]

Thanks for asking the question. There isn't a single elegant solution for this use case. If I were implementing it, I would suggest the following approach:

1. Use a first workflow that runs either an agent or a Responses API call to make changes to the given image and save it on OAI storage.
2. Then, have a second agentic workflow that utilizes your MCP server to process the updated image.

I realize this might not be the ideal answer you're hoping for, but instead of combining everything into a single workflow, I recommend separating the concerns in this way. Also, using a URL to pass images is the best way to go.