comet_screenshot tool not working - CDP captureScreenshot type mismatch

[whoiskjl]

Description

The comet_screenshot tool is not working properly. When called, it fails to return screenshot data.

Technical Analysis

Root Cause

The issue is in src/cdp-client.ts at line 1113:

async screenshot(format: "png" | "jpeg" = "png"): Promise<ScreenshotResult> {
  this.ensureConnected();
  return this.client!.Page.captureScreenshot({ format }) as Promise<ScreenshotResult>;
}

Problems identified:

1. ensureConnected() is not being awaited, which means the connection check happens asynchronously but doesn't block execution
2. The type assertion as Promise<ScreenshotResult> may be hiding a runtime type mismatch
3. No error handling if the CDP method fails or returns unexpected data
4. Missing additional CDP parameters that might be required for reliable screenshots

Expected Type

From src/types.ts line 27:

export interface ScreenshotResult {
  data: string; // Base64 encoded
}

Proposed Fix

Update the screenshot method in src/cdp-client.ts:

async screenshot(format: "png" | "jpeg" = "png"): Promise<ScreenshotResult> {
  await this.ensureConnected(); // Add await
  
  try {
    const result = await this.client!.Page.captureScreenshot({ 
      format,
      captureBeyondViewport: false // Add CDP options
    });
    
    if (!result || !result.data) {
      throw new Error("Screenshot returned no data");
    }
    
    return { data: result.data };
  } catch (error) {
    throw new Error(`Screenshot failed: ${error instanceof Error ? error.message : String(error)}`);
  }
}

Environment

• Platform: Windows
• Version: 2.5.0 (current main branch)

Impact

Users cannot capture screenshots of the current browser state, which is essential for debugging and monitoring agentic browsing tasks.Platform:

[RapierCraft]

Hey @whoiskjl - sorry for the delay, was out for the weekend!

Thanks for the detailed report and for opening a PR to improve the tool - really appreciate it! 🙏

I've looked at the code and had a few thoughts:

Regarding ensureConnected() not being awaited

The ensureConnected() method at line 1184 is actually synchronous - it returns void, not a Promise:

private ensureConnected(): void {
  if (!this.client) {
    throw new Error("Not connected to Comet. Call connect() first.");
  }
}

So there's nothing to await here. You may be thinking of the async ensureConnection() method (line 214) which does different things (health checks + reconnect).

What could actually be the issue

That said, I do see a potential robustness problem: the screenshot() method doesn't use our withAutoReconnect() wrapper that other methods like safeEvaluate() use. This means if the CDP connection drops mid-operation, it won't auto-recover.

Questions for you

1. What error message are you actually seeing? (full stack trace if possible)
2. When does it fail? - immediately, or after some delay?
3. Does comet_connect succeed before you call comet_screenshot?
4. Are you using the HTTP bridge or local stdio transport?

This will help me understand if it's a connection issue, a CDP protocol issue, or something else entirely.

[whoiskjl]

1. no error msg; just no content

2. it doesn't fail - it just returns an empty content. 3. Yes 4. stdio

[RapierCraft]

Thanks for the additional info! That helped narrow it down.

Since you're getting empty content (not an error), the likely cause is the screenshot being captured before the page fully renders. I've put together a fix in #4 that:

1. Waits for page readiness - checks document.readyState === 'complete' before capturing
2. Adds a small render delay - 100ms buffer after load to ensure painting is done
3. Validates the result - throws a descriptive error if data is empty instead of silently returning nothing

I tested it locally and it's working:

5. Testing screenshot...
   Time: 202ms
   Data length: 55116

Screenshot came out as a valid 40KB PNG (1414x804).

Would you mind giving it a try?

git clone https://github.com/RapierCraft/Perplexity-Comet-MCP.git
cd Perplexity-Comet-MCP
git checkout fix/screenshot-empty-data
npm install
npm run build

Then point your Claude Desktop config to the built version. Let me know how it goes!

[whoiskjl]

still the same issue, returned content with comet_screenshot is just empty. Not sure if the issue lies somewhere else.

If you're not experiencing the same issue, I will just a bit more research on my end to see if it's my machine specific.

I will close it for now.

[RapierCraft]

Hey @whoiskjl - sorry to hear the fix didn't help! Since you're still seeing empty content, let me share some additional debugging steps that might help track this down.

Quick diagnostic script

If you have a chance, try running this in Node.js with the browser open:

// Save as test-screenshot.mjs
import WebSocket from 'ws';

const PORT = 9223;
const response = await fetch(`http://127.0.0.1:${PORT}/json/list`);
const targets = await response.json();
const page = targets.find(t => t.type === 'page');

console.log('Target info:', {
  type: page.type,
  title: page.title?.substring(0, 30),
  url: page.url?.substring(0, 50)
});

const ws = new WebSocket(page.webSocketDebuggerUrl);
await new Promise(r => ws.on('open', r));

let id = 1;
const send = (method, params = {}) => new Promise((resolve, reject) => {
  const msgId = id++;
  const handler = (data) => {
    const msg = JSON.parse(data.toString());
    if (msg.id === msgId) { ws.off('message', handler); resolve(msg); }
  };
  ws.on('message', handler);
  ws.send(JSON.stringify({ id: msgId, method, params }));
  setTimeout(() => reject(new Error('timeout')), 5000);
});

await send('Page.enable');
const result = await send('Page.captureScreenshot', { format: 'png' });

console.log('Result:', {
  hasError: !!result.error,
  error: result.error?.message,
  hasData: !!result.result?.data,
  dataLength: result.result?.data?.length || 0,
  dataPreview: result.result?.data?.substring(0, 20)
});

ws.close();

Run with: node test-screenshot.mjs

Things to check

1. Which tab is connected? - Sometimes CDP connects to a DevTools panel or extension page instead of the main page. The type and url in the output above will help confirm.

2. Is it a headless/minimized issue? - Some systems have issues capturing screenshots when the window is minimized. Try with the browser visible and focused.

3. Perplexity version - Are you on the latest Perplexity build? Sometimes updates change the Electron/Chromium internals.

4. Windows version - Any chance you're on Windows 10 vs 11? I've seen some CDP quirks between versions.

Feel free to reopen this if you find more info - happy to dig deeper!