Features
Everything you need to find the bugs you’re missing.
Six interlocking capabilities. Each one does something specific. Together they close the gap between “it works in code review” and “it works in production.”
01 · Exploratory AI
An AI that actually clicks through your app.
Claude drives a real Chromium browser via a local proxy. It reads the page like a human would — understands buttons, menus, forms, and navigation — then makes decisions about what to try next. No fixed scripts, no brittle selectors.
- Reads live page structure and content, not brittle pixel coordinates
- Adapts to dynamic content, modals, and multi-step flows
- Builds a knowledge graph of your app across sessions
- Prioritises unexplored paths over already-visited ones
- Flags when a page state changes unexpectedly mid-session
02 · Local proxy
Your code never leaves your machine.
A small Go daemon runs on localhost. It manages a headless Chromium, handles the WebSocket tunnel to the cloud, and keeps your source code and credentials local. Screenshots and DOM snapshots are the only data that travels.
- Single binary — macOS, Windows, Linux
- Works against localhost, preview deploys, or staging tunnels
- No inbound firewall rules required
- Credentials stored in your OS keychain, with an encrypted-file fallback if none is available
- Loopback tunnel — no VPN or SSH required
03 · Bug inbox
A real bug report, not a stack trace.
Every finding includes what was expected, what happened, the exact steps a human can follow to reproduce, and the AI reasoning trail. Confirm, dismiss, or queue a retest in one click.
- Severity classification: critical, major, minor, or info (the default when unset)
- Reproduction steps you can hand to any developer
- Bulk confirm/dismiss with keyboard shortcuts
- Full audit trail of every status change
04 · Live session view
Watch the AI work in real time.
Every session streams its reasoning, actions, and screenshots as it runs. You see exactly what Claude decided to do and why — pause, inspect, then let it continue. No black box.
- Real-time reasoning stream with token-level updates
- Session timeline with per-step cost breakdown
- Hard stop with graceful budget cap
- Exportable session transcript for post-mortem analysis
05 · Webhooks
Bugs where your team already lives.
Send new findings to Slack, Microsoft Teams, or Discord the moment they’re confirmed. Export to GitHub issues or GitLab issues with one click from the bug drawer. Configurable per-project, per-event.
- Slack Block Kit · Teams Adaptive Card · Discord embed
- GitHub issue export with labels and assignees
- GitLab issue export
- HMAC-signed payloads for security
- Exponential backoff retry on delivery failure
06 · IDE integration
Your coding agent can ask ClickProbe directly.
Expose findings over MCP. Claude Code, Cursor, or your own agent can query the latest run, trigger a retest, or open a GitHub issue — all without leaving the editor.
- MCP server bundled in the proxy — no extra install
- Tools: trigger_test, get_bugs, get_bug_detail, retest_bug
- Works with Claude Code, Cursor, and VS Code Copilot
- Project-scoped — agents only see their own findings