What feature would you like to see?

I would love to see support for MCP tool responses that contain images, according to the MCP specification .
By "support" I mean "let the model see the content of the images". I'm not talking about rendering images in the terminal or anything like this, just have an agent iterating over tools that respond with images (or combinations of images and text).

Additional information

I added experimental_use_rmcp_client = true to the config, and if I use an MCP tool that returns an image it shows something like this in codex:

• Called my-mcp-test.get_image({})
  └ <image content>

However the model can't "see" the image, and the base64 content is consumed as is. I tried to order it write the base64 to a file and run shell commands to decode it to a new .jpg file, but it took minutes of reasoning before it failed.

The only workaround I found was to change my MCP server tool to save the image file, and then use the view_image tool in codex to view the file.

Last time I checked this is also unsupported via the Responses API.