[BUG] Screenshot is not retrieved - Figma Dev Mode

[Sniper199999]

Problem (one or two sentences)

Roo chat is unable to retrieve Figma Screenshot when it calls the "get_screenshot" function.
It returns "(No Response)"
Other functions are able to run correctly...

Context (who is affected and when)

Happens when LLM uses Figma MCP tool "get_screenshot"

Reproduction steps

1. Select a Figma Design/Page in the Figma Desktop
2. In Roo Chat ask to create UI using Figma Design and also fetch screenshot for reference

Expected result

Screenshot is passed to the LLM

Actual result

No screenshot is passed to the LLM

Variations tried (optional)

No response

App Version

v3.41.3

API Provider (optional)

OpenRouter

Model Used (optional)

Gemini 3 flash preview

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

no logs

[roomote]

I've analyzed this issue and identified the root cause.

Root Cause Analysis

The issue is in src/core/tools/UseMcpToolTool.ts. The processToolContent method only handles text and resource content types, but the MCP protocol also supports image content types (which is what Figma's get_screenshot tool returns).

Currently, when an MCP tool returns an image:

// In processToolContent method - lines 253-271
return toolResult.content
    .map((item: any) => {
        if (item.type === "text") {
            return item.text
        }
        if (item.type === "resource") {
            const { blob: _, ...rest } = item.resource
            return JSON.stringify(rest, null, 2)
        }
        return ""  // <-- image types fall through here and return empty string!
    })
    .filter(Boolean)
    .join("\n\n")

When all content items are images, this returns an empty string, which becomes "(No response)".

Interestingly, the access_mcp_resource tool in accessMcpResourceTool.ts (lines 66-80) does handle images correctly by extracting them and passing them to pushToolResult.

Proposed Implementation Plan

1. Modify processToolContent in UseMcpToolTool.ts to return both text content and images separately (instead of just a string)

2. Update executeToolAndProcessResult to:

   • Extract images from the tool result content (items with type === "image")
   • Convert image data to data URLs (handling both raw base64 and already-formatted data URLs)
   • Pass images to task.say() and pushToolResult() just like accessMcpResourceTool does

3. Add appropriate tests to verify image handling in MCP tool responses

The fix will follow the same pattern already established in accessMcpResourceTool.ts:

// Example of handling images (from accessMcpResourceTool.ts)
let images: string[] = []

resourceResult?.contents.forEach((item) => {
    if (item.mimeType?.startsWith("image") && item.blob) {
        if (item.blob.startsWith("data:")) {
            images.push(item.blob)
        } else {
            images.push(`data:${item.mimeType};base64,` + item.blob)
        }
    }
})

await task.say("mcp_server_response", resourceResultPretty, images)
pushToolResult(formatResponse.toolResult(resourceResultPretty, images))

Please mention @roomote in a comment saying that you approve this plan, and I'll proceed with the implementation.

[Sniper199999]

@roomote I approve of the plan, lets fix this

[roomote]

Thanks for the approval @Sniper199999! I've seen your comment and will now implement the fix. I'll open a PR to share progress shortly.

[roomote]

Opened PR #10874 with my attempt at addressing this. Feedback welcome!

The fix adds support for image content types in MCP tool responses, following the same pattern already used in accessMcpResourceTool.ts.