@arwars/simplebug-mcp

MCP server for SimpleBug. It lets an AI agent pull everything captured in a bug report — reproduction steps (incl. keyboard), console logs, network requests (failures flagged), device info, and video frames — so it can diagnose and fix the reported bug.

It's a standard MCP stdio server, so it works in any MCP client: Claude Code, OpenClaw, Antigravity CLI, Cursor, Windsurf, etc.

Requirements

• Node.js ≥ 20
• ffmpeg / ffprobe on PATH (only needed for the video-frame tool)
• A SimpleBug API key — generate your own in the SimpleBug web app under Settings → API Keys. Keys are per-user with per-organization permissions, so you only see reports for the orgs you belong to.

Configuration

| Env var | Required | Default | | --- | --- | --- | | SIMPLEBUG_API_KEY | yes | — | | SIMPLEBUG_URL | no | https://simplebug.example.com |

Install

Claude Code

claude mcp add simplebug -s user \
  --env SIMPLEBUG_API_KEY=sbk_your_key \
  --env SIMPLEBUG_URL=https://simplebug.example.com \
  -- npx -y @arwars/simplebug-mcp

OpenClaw / Antigravity CLI / Cursor / generic MCP client

Add an MCP server entry:

{
  "mcpServers": {
    "simplebug": {
      "command": "npx",
      "args": ["-y", "@arwars/simplebug-mcp"],
      "env": {
        "SIMPLEBUG_API_KEY": "sbk_your_key",
        "SIMPLEBUG_URL": "https://simplebug.example.com"
      }
    }
  }
}

Tools

• get_bug_report(report) — full report: metadata, reproduction steps (incl. keyboard), console logs, network requests (failures flagged), video URL, device. Accepts a share link or report id.
• list_bug_reports(limit?, status?) — browse recent reports you can access.
• get_network_request(report, request_id) — full headers/bodies of one captured request.
• get_bug_report_screenshots(report, at_seconds?) — extract video frames to local image files and return their paths + video time + nearest reproduction step, so you can inspect visual bugs with your own vision tool. Targeted by default (a few orientation frames); pass at_seconds to zoom into a moment. Never decodes the whole video.

Typical flow

1. get_bug_report to read the steps / errors and form a hypothesis.
2. get_bug_report_screenshots with at_seconds at the suspected moment(s); look at the returned frame paths with your own vision tool and compare a few.
3. Fix the bug.