Skip to content

Bug reporting for developers: the evidence you would have collected yourself

Updated

Bug reporting for developers works best when the report contains what you would have collected yourself: the console error with its stack trace, the failing request with its body, the steps that led there and the browser it happened in. Midge captures all of that automatically from a Chrome tab, lines it up with the video on one timeline, and redacts secrets before upload, so the report is safe to paste into a ticket, a pull request or an AI chat.

Every recording includes console logs, network requests, user actions and device info. That means fewer "can you send me the console?" messages and more bugs fixed from the first read.

What you get in a report

One timeline

The video sits next to a panel of console messages, network requests and user actions, all ordered by time. Watch the click on Save and the PATCH /api/projects/88 that followed it is right there, with the Uncaught TypeError a few milliseconds after the response arrived. You do not reconstruct the sequence from three attachments.

Console with stack traces

All levels from debug to error, plus uncaught exceptions and unhandled promise rejections with their stacks. Midge starts listening as the page loads, so startup errors are included, and it keeps a rolling buffer, so errors from before the tester pressed record are still there. Details on the console logs page.

Network detail

Fetch and XHR requests come with headers, request and response bodies, status codes and timings. WebSocket messages are recorded in both directions. Other resources show status, size and timing. For GraphQL, each request shows its operation name, and 200 responses that carry an errors array are flagged, since a status filter would miss them. More on the network logs page.

Device info

Browser and version, operating system, window size, language, time zone and URL. Enough to set up the same environment, or to notice that the bug only happens in a German locale.

Copy as cURL

Any request can be copied as a cURL command and run from your terminal:

curl 'https://staging.acme.dev/api/projects/88' \
  -X PATCH \
  -H 'Content-Type: application/json' \
  -H 'Authorization: [REDACTED]' \
  --data-raw '{"name":"Q4 roadmap","archived":null}'

The redacted Authorization header is intentional. Put in a token for your own test user and you are reproducing the exact request without anyone sharing a live session. If you want to inspect a token from your own environment, the free JWT decoder shows its claims and expiry without sending it anywhere.

Download a redacted HAR

When you need the whole network log, for a performance question or for an API vendor's support team, download it as a HAR file. The file contains the redacted data. Open it in the free HAR viewer, or in any tool that reads HAR. If you are working from a HAR someone exported straight from DevTools, read how to remove sensitive data from a HAR file before you forward it.

Redaction so reports are safe to paste

Bug reports get pasted everywhere: tickets, Slack threads, pull request descriptions, chats with coding assistants. Passwords, card numbers, tokens and cookies are redacted in your browser before anything leaves your device. That includes:

  • Authorization, Cookie and Set-Cookie headers, and any header with token, secret, session or auth in its name,
  • sensitive URL parameters like access_token, code and signature,
  • body fields like password, otp, cvv and client_secret,
  • JWTs, Luhn-valid card numbers, private keys and API keys from Stripe, AWS, GitHub, Slack, Google, OpenAI and Anthropic.

Request IDs and other random-looking values are kept on purpose, because you need them to find the matching server logs. Redaction is rule-based; if your app has its own secret fields, add them to your workspace's list. The full rules are on the redaction page.

Working with coding agents

Midge has an MCP server at https://midge.dev/mcp, so Claude Code, Cursor and other agents can read a report's console, network and actions directly. Paste the report link into the chat and the agent fetches what it needs with read-only tools such as get_console, get_network and get_user_actions. It signs in through your browser, or uses a token from Settings › Account › AI assistants. You can still copy the failing request as cURL or attach the redacted HAR. Either way the data is already scrubbed.

Getting better reports from your team

The best tool still depends on what people capture. Share our guide on how to write a bug report with testers and support, and point them at the screen recorder. Midge is free: unlimited recordings, videos up to 15 minutes, and no watermark, so nobody has to ration recordings.

Frequently asked questions

What does a developer see when opening a Midge report?
The video or screenshot, with console messages, network requests, user actions and device info next to it, ordered by time. Requests show headers, bodies, status codes and timings.
Can I replay a failing request?
Yes. Any captured request can be copied as a cURL command. Auth headers and cookies are redacted, so you add a token for your own test account before running it.
Is it safe to paste a Midge report into a public issue or an AI chat?
Midge removes passwords, card numbers, tokens, cookies and known API key formats before upload. Redaction is rule-based, so review anything unusual, but the common secrets are already gone.
Can I use Midge for my own debugging?
Yes. Recording a bug you found yourself is a quick way to hand it to a teammate or come back to it later with the logs intact.