Blog ·
Open sourcing the Webtrax runner
The Webtrax runner is open source under the Apache-2.0 licence. It is the engine the Webtrax agent will run on, and it is useful on its own today.
npx @webtrax/cli check https://your-site.example
What it checks
- Links: broken internal and external links, timeouts, domains that no longer resolve, redirect chains, HTTPS pages linking to HTTP, stale sitemap entries, orphan pages, “not found” pages that return 200, and canonical links that redirect or fail.
- Security, passively: HTTPS and its enforcement, HSTS, Content-Security-Policy (missing or unsafe), the other security headers, mixed content, insecure forms and version disclosure. It reads only what a browser visit would.
- SEO basics: titles, meta descriptions, headings, noindex left over from staging, and page language.
- Accessibility and performance, with
--browser: axe-core’s WCAG 2.x A and AA rules, and Largest Contentful Paint, layout shift, time to first byte and page weight, measured in Chromium on a sample of pages. The browser goes through the same network guard as the crawler.
Each finding has a stable id, the evidence, and how to fix it. Reports come as a terminal summary, Markdown for pull requests or for pasting into an AI chat, JSON, SARIF for GitHub code scanning, JUnit for CI dashboards, and a single-file HTML report. Every rule has a page in the docs.
For CI
There is a GitHub Action (coldbathcode/webtrax-runner/action@v1), and exit codes are built for pipelines: 0 when clean, 1 when a finding is at or above --fail-on (high by default), 2 for a usage error and 3 when the site cannot be reached. webtrax diff compares two runs, so you can see what a change added and what it fixed.
For coding agents
webtrax mcp is a local MCP server. Add it to Claude Code with claude mcp add webtrax -- npx -y @webtrax/cli mcp, or to Cursor and any other MCP client, and your agent can check your dev server before it commits: audit a page, check the links it just wrote, or crawl the site and list only what is new since the last run. It checks local addresses only unless you start it with --allow-public, and it marks everything read from web pages as untrusted, so a page cannot instruct your agent.
A polite crawler
The runner identifies itself as WebtraxBot, honours robots.txt and Crawl-delay, slows down when a site rate-limits it, and refuses private networks and cloud metadata addresses unless you ask for them. More at webtrax.dev/bot.
The source is at github.com/coldbathcode/webtrax-runner. The hosted Webtrax agent, which checks every preview deploy and opens fix pull requests, comes next. Join the waitlist for early access.