[Bug] - Images returned by MCP tools are not displayed in chat or accessible as files

[AlonDan]

Describe the bug
When any MCP tool returns image data (such as base64-encoded PNG), Zoo Code does not display the image in chat and does not make the base64 data accessible to the AI model.

This is a Zoo Code bug, not an MCP bug. The MCP server correctly returns the image data, but Zoo Code fails to:

1. Display the image in the chat interface
2. Make the base64 data available to the AI model

I verified this with the game_screenshot tool from godot-mcp:

• The tool returns success: true with dimensions (e.g., "1152x648")
• The actual image data is returned but I cannot see it or access it
• No PNG file is created in the project directory

This bug affects ALL MCP tools that return image data, not just game_screenshot.

To Reproduce
Steps to reproduce the behavior:

1. Install godot-mcp MCP server (version 3.1.0)
2. Run a Godot project using run_project tool
3. Call game_screenshot tool
4. Expected: See the screenshot image in chat or access the base64 data
5. Actual: Only see text message "Screenshot captured: 1152x648" — no image

Expected behavior
When an MCP tool returns image data, I expect:

• The image to be displayed in the chat interface
• OR the base64 data to be accessible so I can save it as a file
• OR a file to be automatically created in the project directory

Screenshots
N/A - I cannot provide screenshots because this is the bug being reported.

Video
N/A - Cannot record video of the bug because screenshots are not displayed.

What version of zoo are you running
Version: Latest (installed via VS Code extension marketplace)

Additional context

• OS: Windows 11 Pro
• MCP server: godot-mcp version 3.1.0
• The game_screenshot tool is enabled in mcp_settings.json (listed in alwaysAllow)
• I verified that other MCP tools (create_scene, add_node, run_project, read_scene, create_script) work correctly
• The bug is specifically with how Zoo Code handles image data returned from MCP tools
• This is a critical bug for Vibe Coding workflow where visual verification is essential

[taltas]

@zoomote Can you investigate why we are not showing images in the chat? If there are no underlying reasons, please attempt a fix and make a PR.

[zoomote]

I'm on it. I started a task for this issue, and I'll keep updates here.

See task

[zoomote]

Investigated and fixed — draft PR: #1025

Root cause (two gaps):

1. Chat UI never rendered the images. The backend already attached image data URLs to the mcp_server_response chat message, but the webview's ChatRow had no case for that message type, so only the [N image(s) received] placeholder text was rendered and the images were silently dropped from the UI.
2. Embedded resource image blobs were discarded. For MCP results that return images as embedded resources, processToolContent stripped the blob before the data could reach the chat or the model.

Fix:

• ChatRow now renders images attached to mcp_server_response messages inline in chat (same ImageBlock pattern used for assistant messages with images).
• Image resource blobs are extracted into data URLs, so they're displayed in chat and forwarded to the model as image blocks (mirroring the existing access_mcp_resource behavior).

One note: images are only sent to the model when the active model profile declares image support (supportsImages) — that capability gating is pre-existing and intentional, so for visual verification workflows make sure you're using a vision-capable model.

Validation: full pnpm test (7010 tests), lint, and type checks pass, plus new unit and webview component tests covering both fixes.

I'm doing a self-review on GitHub and will follow up here with those results.