[FEATURE] Let the agent read image files with its own vision capabilities

[ZeterMordio]

Claude Code has vision. It can analyze screenshots, describe UIs, spot visual bugs. But it can only use this ability when I manually paste an image into the chat (Cmd+V). The agent itself has no way to look at an image file on disk.

This means that even though Claude can take a screenshot via screencapture or generate a chart via code, it can't actually look at the result. The Read tool returns "Binary files are not supported" for images. There is no ReadImage tool. The agent is effectively blind.

Why this matters

The obvious use case is UI development. Right now, if I ask Claude Code to build a frontend component, it writes the code and hopes for the best. It can't take a screenshot of the result and check if it looks right. A simple ReadImage tool would close that loop and let the agent iterate visually without me having to paste screenshots back and forth.

This generalizes to anything visual: comparing a rendered page against a design mockup, catching layout regressions, verifying chart output, debugging "it looks wrong" issues.

What I'd like to see

Extend the Read tool to handle image files (PNG, JPEG, WebP, etc.) by feeding them into the model as vision input, the same way a pasted image works today. The vision model already supports this. The gap is just in the tool layer.

Context

• No CLI coding agent has fully solved this. Gemini CLI has the same gap (Is it possible to make Gemini-cli support image(screenshot) input? google-gemini/gemini-cli#3596, Screenshot into CLI google-gemini/gemini-cli#14474). OpenAI's cloud-based Codex can do it (it spins up a browser and views its own work), but the local Codex CLI cannot.
• Community workarounds exist (Playwright MCP, round-trip test harnesses, Chrome DevTools MCP), but they're all browser-specific and heavyweight. None of them solve the general case of "read this image file."
• Several existing issues touch on parts of this problem: [BUG] - Read tool: Claude cannot interpret or see image content when reading image files #18588 (Read tool can't interpret images), Title: Add screen capture and variable inspection capabilities for debugging to Claude Code #20114 (screen capture for debugging), Add Puppeteer support for taking browser screenshots #4806 (Puppeteer screenshots), Screenshot Support via Clipboard Paste in Claude Code CLI #12644 (clipboard paste support). This is the underlying capability they all need.

[github-actions]

Found 3 possible duplicate issues:

1. [BUG] - Read tool: Claude cannot interpret or see image content when reading image files #18588
2. Allow Claude to see images by itself #592
3. [BUG] Claude cannot read images anymore. #560

This issue will be automatically closed as a duplicate in 3 days.

• If your issue is a duplicate, please close it and 👍 the existing issue instead
• To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

[ZeterMordio]

Not a duplicate of any of these three:

• [BUG] Claude cannot read images anymore. #560 was a regression where image viewing broke entirely, even with paste. It was fixed and closed by an Anthropic engineer. Different problem.
• Allow Claude to see images by itself #592 asked for the same base capability back in March 2025 but got auto-closed by the stale bot with 2 upvotes and no response. It's now locked.
• [BUG] - Read tool: Claude cannot interpret or see image content when reading image files #18588 is a narrower bug report about the Read tool returning nothing for image files.

This issue consolidates the ask from all of them and explains why it matters: the agent has vision capabilities but no tool-level pathway to use them autonomously. That's the gap that #18588, #20114, #4806, and #12644 all run into from different angles.

[github-actions]

Closing for now — inactive for too long. Please open a new issue if this is still relevant.

[github-actions]

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.