How tools should be created to use images in Claude Desktop?

[alexlopezcifuentes]

So basically, I am trying to follow the simple quickstart guide on how to work with images.

from mcp.server.fastmcp import FastMCP, Image
from PIL import Image as PILImage

mcp = FastMCP("My App")


@mcp.tool()
def create_thumbnail(image_path: str) -> Image:
    """Create a thumbnail from an image"""
    img = PILImage.open(image_path)
    img.thumbnail((100, 100))
    return Image(data=img.tobytes(), format="png")

However, it does not seem straigthforward how to use this in a conversation with Claude. I have properly configured the MCP server in Claude, and those tools not using images are properly working.

Is there any way of uploading an image to Claude so that it is then used by an MCP tool?

[llermaly]

+1

[tjroamer]

Is there solution to this issue?

[deeplook]

I'm also running into this. When using and even simpler code snippet than the default thumbnail example I get either errors (and not having a Pro accounts seems to make them even more unpredictable), or when I skip using PIL I'll get a textual description of the image, which is nto what I expected.

@mcp.tool()
def show_image(image_path: str) -> Image:
    """Show an image"""
    # This will raise an error: "Error executing code: Maximum call stack size exceeded"
    img = PILImage.open(image_path)
    return Image(data=img.tobytes(), format="png")
    
    # this will return a textual description of the image!
    # return Image(path=image_path, format="png")

[deeplook]

In fact, I'm not sure if FastMCP inside this SDK is developped any further. If you take a similar example from FastMCP 2.0 it works.

[slovx2]

I’m encountering the same issue.
When using this SDK to return an Image, most clients can’t recognize it (they treat the base64 as plain text and pass it directly to the agent).
Clients I’ve tried: Claude Desktop, Claude Code, Cherry Studio