Can I give Claude code screenshots?

2026-10-13

Illustration for “Can I give Claude code screenshots?”

Yes—if your Claude Code setup can access and inspect images. You can provide a screenshot as visual evidence, but a screenshot alone shows one moment. If the bug depends on a sequence of actions—clicking a button, changing a setting, or moving through a flow—a recording with timestamps and a short explanation can give Claude Code more context.

The important distinction is between having an image and giving the agent a way to read it. Claude Code needs an image-attachment or other image-inspection capability in your setup. If you share a link instead, the agent needs tools and permissions that let it retrieve the linked content and inspect the image. A URL by itself does not guarantee access.

When a screenshot is enough

A screenshot can be a good fit for a static visual issue: a misaligned button, a clipped heading, an unexpected error message, or a layout that looks wrong at a particular screen size. It gives the agent a view of what was visible at that moment.

Pair the image with a concise description of the expected and actual behavior. For example:

The Save button is visible, but clicking it leaves the old name on the page. I expected the new name to appear. Please inspect the relevant code and suggest what to check. Don’t change anything until you explain your reasoning.

That description is a template, not a claim about a particular bug. Adapt it to the behavior you observed. It helps separate your expectation from what the image actually shows—and gives the agent a specific question to investigate.

A screenshot does not, by itself, show what happened before or after it was captured. If the issue is caused by a click, a state change, or a particular order of steps, add those details in text or provide a recording.

When to use a screen recording instead

Use a recording when the sequence matters. A dropdown that closes unexpectedly, a page that changes after a click, or a flow that fails only after several steps can be difficult to explain with one still image.

A useful recording handoff includes:

The steps you took before the problem appeared.
What you expected to happen.
What happened instead.
The moment in the recording where the behavior is visible.

Keep the capture focused on one issue. Before sharing, review what is visible and audible: private account details, customer data, keys, or unrelated tabs may disclose information you did not mean to send.

How to give Claude Code a ThinkRun recording

ThinkRun’s published workflow uses the recording’s accompanying evidence rather than relying on Claude Code accepting a raw video upload. The workflow makes a video, transcript, screenshots, and timestamps available; Markdown and JSON provide ways to read the accompanying information. The walkthrough demonstrates recording and sharing—not an agent diagnosing or fixing a bug. You still need to review Claude Code’s analysis and verify any proposed change yourself.

A practical handoff looks like this:

Record the relevant workflow. Select the screen, window, or tab you want to capture. Explain what you expected and what happened instead as you reproduce the issue.
Review the evidence. Check the transcript and screenshots. Confirm that the relevant moment is present, and compare any screen interpretation with the actual image.
Choose what to share. Use the replay page when a person needs to watch. Markdown can be a readable starting point for an agent; JSON may be useful when your tools need structured fields. Check the output before sharing it.

Give Claude Code a clear request. Paste the Markdown share link and tell it what to inspect. For example:

Read this ThinkRun recording’s Markdown: <paste the Markdown share link>

Inspect the linked screenshots, especially the moment I describe the problem.
Summarize what happened and cite the relevant timestamps.
Separate observed facts from possible causes.
Then inspect this project’s code and suggest what to check next.
If you cannot open a link or image, tell me what you could not access.

Check the response before acting. Look for a summary tied to specific evidence, not a confident guess. If Claude Code cannot retrieve the shared content or inspect images, use the file and image attachment tools available in your setup instead.

This is not a guarantee that every Claude Code configuration can fetch a shared link. The agent needs access to retrieve the content and inspect the linked images; what it can do depends on the tools and permissions in that session.

What to ask Claude Code to report

Ask it to distinguish observations from hypotheses. For instance, “the menu closes after the click” is an observation if that is visible in the evidence. “The click handler is causing the problem” is a possible explanation that still needs to be checked against the code.

You can also ask Claude Code to cite the relevant screenshot or timestamp, list what it could not verify, and suggest the next code area to inspect. Review that summary before asking it to modify files. Then reproduce the original steps again to check whether the behavior changed.

So, can you give Claude Code screenshots?

Yes, when your session can inspect the image. For a static visual problem, a screenshot plus a precise description may be enough to start. For a bug that unfolds across several steps, a recording with narration, screenshots, and timestamps gives the agent more context to examine. In either case, make sure Claude Code can actually access the evidence—and treat its explanation as something to verify, not as proof of the cause.

To try a recording-based handoff, visit ThinkRun.