Edition II · AI-native · self-hosted · MCP-ready

The bug-reporting tool
built for the era
of AI coding agents.
Self-hosted.

Every report is triaged by Claude — severity, type, duplicates, repro steps — before you open it. Replay → Playwright turns a session recording into a runnable test. The built-in MCP server pipes bugs straight into Claude Code, Cursor, and Windsurf so an agent can fix them with full context. Your data never leaves your infra.

Spin it up in 5 minutes → See the AI features →
Claude triage on every report MCP for Claude Code & Cursor Free for individuals · GDPR by design
bugs.yourco.com
Reports
Triage 3
Projects
Integrations
Settings
All
New
In Progress
Checkout button unresponsive on iOS Safari
2 min ago · Safari 17 · sev 4
MAJOR
Avatar upload silently fails for HEIC files
1 hr ago · Chrome 131 · sev 3
FUNC
Dark mode toggle resets on page reload
3 hr ago · Firefox 126 · sev 2
MINOR

Observed in the wild on

· · · · · ·

10k+

Reports triaged daily

< 30s

From submit to triaged

100%

Self-hostable

GDPR-ready SOC 2 roadmap AES-256 at rest
Chapter 00 · the differentiator

Built for the era of AI coding agents.

Most bug trackers are databases with a UI bolted on. BugPilot treats every report as something an AI can read, reason about, and act on — natively, end-to-end.

flagship № 01

MCP server, built-in.

Plug BugPilot directly into Claude Code, Cursor, Windsurf, and VS Code via the official Model Context Protocol. Your agent gets first-class access to bug context — no copy-paste, no screenshots in chat, no losing thread. 

// ~/.config/claude-code/mcp.json
{
  "mcpServers": {
    "bugpilot": {
      "transport": {
        "type": "http",
        "url": "https://bugs.you.com/api/v1/mcp",
        "headers": { "Authorization": "Bearer bpk_…" }
      }
    }
  }
}
  • bugpilot_search_reports · filter by severity, status, project, full-text
  • bugpilot_get_report · full payload incl. AI analysis, pinned element, replay link
  • bugpilot_get_replay · rrweb event array for the agent to reason over
  • bugpilot_get_playwright_test · runnable test that reproduces the bug
  • bugpilot_update_status · agent moves the report through the workflow
  • bugpilot_add_comment · agent leaves a note when it's done
  • bugpilot_assign_tags, bugpilot_set_known_issue, bugpilot_get_similar

Two scopes — mcp for read access, mcp:write for mutations. Independent. Self-hostable. Bug data never leaves your infra (counter-positions Jam.dev, which is cloud-only).

automated № 02

AI triage on every report.

Claude Haiku-4-5 reads the screenshot, console errors, network failures, and the user's description, then writes back: severity (1–5), type, clean title, repro steps, duplicate detection, and spam filtering. Async via BullMQ — your dashboard updates live via SSE.

on-demand № 03

Fix suggestions with code context.

Click "Suggest a fix" on any report. Optionally paste the relevant code. Claude returns root cause, affected areas, and a 2-3 sentence actionable fix — with a confidence badge. Cached on the report so re-fetches don't re-prompt.

replay → test № 04

Replay → Playwright export.

The widget records the last 30 seconds of user interactions as rrweb events. We extract a clean action timeline (clicks, inputs, navigations with selectors) and ask Claude to translate it into a runnable .spec.ts. Drop it into your test suite — passive evidence becomes an active regression test.

precision № 05

📍 Pin the element.

The reporter clicks any element on the page; the widget records its CSS selector + XPath + bounding box (page-coordinate, scroll-aware). The selector flows into the AI repro steps — your devs jump straight to the broken DOM node in DevTools, no archaeology required.

privacy № 06

PII redaction before AI sees it.

Per-project, opt-in. Server-side scrubber runs on ingest, before the report hits Claude or your storage. 15 categories — emails, credit cards (Luhn-validated), JWTs, API keys, AWS keys, IBANs, crypto wallets, PEM private keys, DB connection strings with embedded credentials, and more.

Self-hosted · your Claude API key · your data, your servers

No middleware vendor between your bugs and your AI. Bring your own CLAUDE_API_KEY and own the whole pipeline.

Side by side

How BugPilot compares.

BugPilot Marker.io Bird Eats Bug Userback
AI triage (severity, dupes, repro steps) ✓ ✗ ✗ ✗
Self-hostable ✓ ✗ ✗ ✗
Annotated screenshots ✓ ✓ ✓ ✓
MCP server (Claude Code / Cursor / Windsurf) read + write tools ✗ cloud-only ✗
Replay → Playwright export ✓ ✗ ✗ ✗
AI fix suggestions with code context title-only rewrite ✗ ✗
Session replay ✓ ✗ ✓ ✗
Server-side PII redaction 15 categories, per-project ✗ ✗ ✗
GitHub / Linear / Jira ✓ ✓ ✓ ✓
Free tier free for individuals 14-day trial 14-day trial 14-day trial
Vendor lock-in none — your data, your infra SaaS only SaaS only SaaS only

Field reports

What teams say.

"AI triage saved our QA lead at least an hour a day. Severity is right ~9 times out of 10."
— Engineering lead, B2B SaaS · 14-person team
"Self-hostable was non-negotiable for us. BugPilot was the only one that didn't make us compromise."
— Platform engineer, fintech
"We were drowning in dupes. The AI flags them on intake — our triage backlog dropped 40%."
— Product manager, indie dev tool

Hands on

There's a real bug button
on this page.

Click the floating button in the corner to open the actual BugPilot widget. Capture a screenshot, draw on it, write a report. See exactly what your team would see — minus the dashboard, since this is just a demo page.

  • One-line embed. <script src="/bugpilot.js" data-key="…"></script>
  • Annotations stored as SVG, not flattened into the screenshot.
  • Console errors and network failures captured automatically.

No signup. No data leaves your browser unless this page has a live API endpoint configured.

Security & compliance

Built for enterprise.

Data residency

Self-host on your own infrastructure. Your reports never leave your network unless you choose otherwise.

Encryption

AES-256 at rest for credentials and TOTP secrets. RS256 JWT, HTTPS everywhere.

GDPR-ready

User data export, account deletion with 30-day grace, DPA template available on request.

SOC 2 roadmap

Working toward SOC 2 Type I in 2026. Audit log, RBAC, and observability shipped.

# specimen BP-26-0420-001
CAPTURED
title: "Checkout unresponsive on iOS Safari"
screenshot: annotated.png // red arrow + blur
console: [TypeError: o.then is not a function]
network: POST /checkout → 500
AI TRIAGED
severity: 4 / major
type: functional
duplicate_of: null // 0.84 confidence
repro: ["Open cart", "Tap 'Buy'", "Nothing happens"]
status: → new
Chapter 01 · anatomy

Every report, fully prepped before you read it.

Regular bug reports are "it's broken, please help". BugPilot reports are a specimen slide — by the time a human looks at one, the context is already there.

  • Screenshot is captured, annotated, and stored on your infrastructure.
  • Console errors and failing network calls are attached automatically.
  • Claude writes a clean title, extracts repro steps, flags duplicates.
  • Severity and type classify the report before it hits your inbox.
Chapter 02 · specimens

Everything you need. Nothing you don't.

Twelve features, zero fluff. Built for small teams who want professional bug tracking and first-class AI workflows without the enterprise price tag.

№ 001 · AI · flagship
🤖

AI triage, automatically

Claude classifies every report — severity (1–5), type, a clean title, reproduction steps, and whether it's a duplicate — all before you open it.

№ 002 · MCP · flagship
🔌

Native MCP server

9 tools (5 read + 4 write) for Claude Code, Cursor, Windsurf, VS Code. Self-hostable. Independent mcp / mcp:write scopes. Replaces copy-pasting bug context into chat.

№ 003 · AI
🛠️

Fix suggestions

Claude reads the report + optional code snippet you paste, returns root cause + actionable fix + confidence. Cached per-report. Available via the dashboard and as an MCP tool.

№ 004 · AI · unique
🎬

Replay → Playwright

Captured rrweb session becomes a runnable .spec.ts. Drop it in your test suite to repro the bug end-to-end. None of the visual bug tools do this.

№ 005 · capture
📍

Pin the element

Reporter clicks the broken element; widget records its CSS selector + XPath + bbox. The selector flows into AI repro steps — devs jump straight to it in DevTools.

№ 006 · privacy
🛡️

PII redaction

15 categories scrubbed server-side before storage: emails, credit cards (Luhn-validated), JWTs, AWS keys, IBANs, crypto wallets, PEM private keys, DB connection strings, and more. Per-project, opt-in.

№ 007 · capture
📸

Annotated screenshots

Users draw, highlight, blur, arrow, rectangle, circle directly on their screen before submitting. Annotations stored as SVG, not flattened into the screenshot.

№ 008 · capture
🎥

Session replay

Opt-in 30s rolling rrweb buffer with periodic FullSnapshots. Inputs masked at capture time. Replayer + scrubber in the report detail. Same capture powers Replay → Playwright.

№ 009 · install
🧩

Chrome / Firefox extension

Bug-report on any site without an embed. QA teams, agencies testing client sites, support reps — load unpacked or sideload via web-ext. Reuses the same widget IIFE.

№ 010 · integrations
🔗

GitHub · Linear · Jira

Two-way sync via webhooks (HMAC-verified). Auto-create issues after AI triage based on per-project severity threshold. Estimate endpoint shows you exactly what would have triggered last 30 days.

№ 011 · realtime
⚡

Real-time dashboard

SSE push updates the UI the moment AI finishes processing. New reports, comments, status changes appear live. Keyboard-first triage inbox: J/K/Space/A/D bulk-triage.

№ 012 · hosting
🏠

Self-hosted, always

Docker Compose, five services, up in under five minutes. 2FA, SSO (OIDC per-org), audit log with tamper-evident hash chain, GDPR export + soft-delete, scoped API keys — all on the free tier.

Chapter 03 · method

From report to fix, in four moves.

Zero friction for your users. Zero noise for your team. No ceremony in between.

i

Drop in the widget

One script tag in your HTML. Pass your widget key and optionally pre-fill reporter info for logged-in users.

<script src="/bugpilot.js" data-key="proj_abc"></script>
ii

User files a report

They click the floating button, annotate the screenshot, type a description, hit send. Under thirty seconds, start to finish.

iii

AI triages instantly

Claude analyses the report, assigns severity, writes a clean title and summary, extracts repro steps, checks for duplicates.

iv

Your team acts

Dashboard updates live. Approve from the triage inbox, push to GitHub or Linear, mark resolved — all in one place.

Chapter 04 · pricing

Simple, honest pricing.

Free for individuals and open-source projects. A commercial license is required if you use BugPilot in a for-profit product.

Personal
Free forever

For individuals, hobbyists, and open-source projects. No license key, no phone-home, no upgrade prompts.

  • 100 AI-triaged tickets / month
  • Unlimited projects
  • GitHub & Linear & Jira integrations
  • Annotated screenshots & replay
  • Self-hosted — your data, your server
  • For-profit commercial use
Get started free
Most popular
Pro
$29 / month

For companies shipping commercial software. Metered billing — pay only for what you use beyond the included quota.

  • 5,000 AI-triaged tickets / month included
  • $0.01 per ticket beyond quota
  • Commercial use license
  • SSO (Google / Microsoft / OIDC)
  • Audit log + 2FA + RBAC
  • Priority email support
Get a license →
Enterprise
Custom

For larger orgs with strict compliance needs, custom integrations, and an SLA.

  • Unlimited tickets — no overage
  • Unlimited team members & orgs
  • Dedicated onboarding & SLA
  • Custom integrations
  • SOC 2 evidence pack on request
  • Invoice billing & DPA
Contact us

Solo developer on a side project? It's free. Company shipping a product to customers? You need a commercial license. Still unsure — hello@bugpilot.dev.

Self-hosted · 5 min

Your infrastructure.
Your rules.

BugPilot ships as a Docker Compose stack — MongoDB, Redis, MinIO, the API, and the dashboard, wired together, health-checked, production-ready.

  • Named volumes for data persistence across upgrades
  • Nginx with gzip, asset caching, and SSE-aware proxy
  • RS256 JWT auth — no shared secrets
  • MinIO for screenshot storage (S3-compatible)
View on GitHub Read the docs
~/bugpilot ● zsh 
# Clone and configure
git clone https://github.com/your-org/bugpilot
cd bugpilot
cp .env.example .env  # fill in secrets

# Generate RS256 key pair
openssl genrsa -out private.pem 2048
openssl rsa -in private.pem -pubout -out public.pem

# Start all services
docker compose -f infra/docker-compose.yml up -d

✓ mongo      healthy
✓ redis      healthy
✓ minio      healthy
✓ api        started on :3000
✓ dashboard  live on :80
ship it.

Stop guessing what broke.

Your first bug report — fully triaged — in under ten minutes from now.

Star on GitHub Spin it up now →