FeedBugSign in

Model Context Protocol

Your agent reads the bug, not your summary of it.

A tester clicks on what is broken. FeedBug captures the screenshot, the console, the failed requests and the DOM around that exact element. The MCP server hands all of it to Claude Code, Cursor or Windsurf, with the likely source file already attached. 

npx @feedbug/mcp init

MIT licensed, published on npm as @feedbug/mcp.

Seven tools, read and write.

feedbug_list_bugsEvery open report on the project, newest first.
feedbug_get_bugOne report with its comment, page URL, status and screenshot.
feedbug_get_diagnosticsConsole logs, JavaScript errors, failed requests, device and browser.
feedbug_get_html_contextThe CSS selector, the component path and the DOM around the element that was clicked.
feedbug_get_replayThe 30 seconds of session replay recorded before the report.
feedbug_add_commentReply on the issue from the editor.
feedbug_resolve_bugMark the bug resolved once the patch is in.

Configure it by hand.

The installer writes this for you, but any MCP client takes the same entry. The project key comes from your FeedBug dashboard and scopes the server to that project alone. 

{
  "mcpServers": {
    "feedbug": {
      "command": "npx",
      "args": ["-y", "@feedbug/mcp"],
      "env": {
        "FEEDBUG_PROJECT_KEY": "pk_your_project_key"
      }
    }
  }
}

Frequently asked questions

What is an MCP server for bug tracking?

MCP, the Model Context Protocol, is how an AI coding agent reads data it does not have in your repository. A bug tracking MCP server exposes your bug reports as tools the agent can call, so it can pull a report and its diagnostics without you copying anything into the chat.

Which clients does it work with?

Any MCP client. FeedBug ships an installer for Claude Code, Cursor, Windsurf and VS Code, and the config is a standard mcpServers entry you can write by hand for anything else.

What can the agent actually do with it?

Seven tools: list open bugs, read one report, pull its diagnostics, pull the DOM context around the element that was clicked, fetch the session replay, comment on the issue and mark it resolved. Reading and writing, so the agent closes the loop without leaving your editor.

Why not just paste the bug into the chat?

Because what matters is not the sentence the tester wrote, it is the CSS selector, the component path, the failed network call and the console error that came with it. Pasting loses all of that. The MCP server hands the agent the structured report, and FeedBug adds the likely source file to it.

Does the agent get access to my whole account?

No. The server authenticates with a single project key, so it only ever sees the bugs of that project.

Is it open source?

The MCP server is published on npm as @feedbug/mcp under the MIT licence.

Point at the bug. Ship the fix.

The MCP server is on the Pro plan. Start free, connect Linear or GitHub, and turn it on when you want your agent in the loop.

Get started

Curious how the source file is found? Read the write-up