squirrelscan

[squirrelscan]

squirrelscan

The website QA tool for your coding agent

squirrelscan is an Open Source cli tool that audits websites for SEO, performance, security, accessibility, agent experience and other issues, and gives your coding agent exact fixes. Run it from the CLI, inside your coding agent, in the cloud, or over MCP.

Combine your coding agent with a deterministic and extensible audit tool.

[Add to Cursor] [Add to Claude Code] [Add to Codex] [Add to opencode] [MCP Registry]

[CI] [CodeQL] [npm] [License: MIT]

[Terminal demo: squirrel audit on a small test site reports 13 failed checks, the report is piped to Claude Code, which fixes the pages and re-audits from 32/100 (F) to 92/100 (A), then git status lists the files it changed]

A real run on a tiny test site, with the released CLI (signed out, so free) and Claude Code. We cut the two minutes the agent spent working, because nobody wants to watch an agent think.

Features

• Built for coding agents - Skills for Claude Code, Cursor, Codex and any coding agent run the audit, fix and re-audit loop. Connect over the CLI, the hosted or local MCP server, or the API.
• Enriched fixes for you and your coding agent - Every finding says how to fix it, in LLM-optimized output your agent can apply, and every rule has its own docs page. Signed in, the report adds an editor's summary of what to fix first and the technology stack the site runs on.
• 300+ rules, 22 categories - Audit coverage across SEO, accessibility, performance, security, content and more.
• Agent experience - Audit how well AI agents can read, discover and act on your site: llms.txt, AGENTS.md, Markdown responses, and what AI crawlers are served.
• Link checking and monitoring - Perch, the cloud link checker, tells a broken link from a blocked one and confirms a failure before reporting it. On Pro and Team, scheduled audits email you when a link breaks.
• Security and site integrity - Detect phishing kits, injected pages, leaked credentials and missing security headers.
• Entity map - Your site's JSON-LD collapsed into one graph: one identity, or a new one on every page.
• Browser rendering in the cloud - Signed in, pages are rendered in a real browser in the cloud, so single-page apps and JavaScript-heavy sites are audited as visitors see them, at no extra cost per page.
• Cloud audits - AI page analysis, scheduled audits, a per-website issue tracker and shareable reports.
• Smart crawling - Fast and memory efficient, incremental with ETag, Last-Modified and content hashing, and resumable from checkpoints.
• Crawl history and changes - Track site evolution, compare crawls, spot regressions.
• Developer-first CLI - Single binary, zero dependencies, shell completions, self-update.
• Multiple output formats - Console, JSON, HTML, Markdown, Text, LLM, XML.

Rule categories

Ordered by how much a failure usually costs you, not by how many rules each one has.

Total: 300+ rules across 22 categories

See the rules reference for full details.

CLI

Installation

Paste into your coding agent (Claude Code, Codex, Cursor, opencode and friends):

Fetch and follow the setup instructions for squirrelscan at https://docs.squirrelscan.com/agent-setup/prompt.md

Your agent installs the CLI, connects MCP and runs a first audit. You can read exactly what it will follow first.

macOS / Linux:

curl -fsSL https://install.squirrelscan.com | bash

Windows:

iwr -useb https://install.squirrelscan.com/install.ps1 | iex

Homebrew (macOS):

brew install --cask squirrelscan/tap/squirrel

The cask installs a launcher that runs the regular installer on first use. The CLI then updates itself with squirrel self update; brew upgrade leaves it alone.

npm (all platforms):

npm install -g squirrelscan

npx (run without installing):

npx squirrelscan audit example.com

Quick start

# Audit a website
squirrel audit example.com

# Generate HTML report
squirrel audit example.com -f html -o report.html

# Pipe to Claude for AI analysis
squirrel audit example.com --format llm | claude

# Quick audit for a fast first look (other levels: surface, full)
squirrel audit example.com --level quick

# run only agent experience and performance rules
squirrel audit example.com --rule-include ax,performance

# login for cloud audits and cloud rendering
squirrel auth login

Reports

A health score and grade, group scores for SEO, performance, security and Agent Experience (AX) colored by score, a ranked list of issues, and a copy fix prompt for each one that you can paste into your coding agent. Publish reports to the web to share with team members or coding agents (fix instructions are embedded).

[A squirrelscan report for squirrelscan.com. The header shows the site, 220 of 220 pages scanned, the Full audit level and a health score of 79 (C). A notice says 912 of 1412 external links were not checked. Four group cards show SEO 86 (B), Performance 59 (F), Security 95 (A) and Agents 42 (F), each with its error and warning counts. Below are pass, warning and fail totals, an auto-generated summary, and the first three ranked issues, each with a Copy fix prompt button.]

See an example report, or read how to read your report.

Add to your coding agent

squirrelscan ships as an MCP server (hosted at mcp.squirrelscan.com), skills (autonomous audit + fix workflows), and a plugin for Claude Code and Cursor. Cursor installs in one click from the badge above; the rest are a single copy-paste.

Cursor

Click the Add to Cursor badge above, or add it manually to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "squirrelscan": { "url": "https://mcp.squirrelscan.com/mcp" }
  }
}

Skills: squirrel skills install

Claude Code

Install the plugin (bundles skills + the MCP server):

/plugin marketplace add squirrelscan/skills
/plugin install squirrelscan@squirrelscan

Or add just the MCP server:

claude mcp add --transport http squirrelscan https://mcp.squirrelscan.com/mcp

OpenAI Codex

Add the server to ~/.codex/config.toml:

[mcp_servers.squirrelscan]
url = "https://mcp.squirrelscan.com/mcp"

Codex reads Agent Skills from ~/.agents/skills, so skills work too: squirrel skills install

opencode

Add the server to opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "squirrelscan": {
      "type": "remote",
      "url": "https://mcp.squirrelscan.com/mcp",
      "enabled": true
    }
  }
}

Any MCP client

squirrelscan is in the MCP Registry as com.squirrelscan/squirrelscan. Point any client at the remote server:

https://mcp.squirrelscan.com/mcp

Authentication is per-user OAuth (or pass a squirrelscan API key as a Bearer token). Skills follow the Agent Skills standard: squirrel skills install puts them in ~/.agents/skills/ and ~/.claude/skills/.

Skills

Two skills drive agent workflows. They live in squirrelscan/skills, along with the Claude Code, Cursor and Agent Plugins packages that bundle them with the MCP server. squirrel skills install downloads and verifies them itself, and from v0.0.99 squirrel's auto-update keeps them current. Installs made earlier with npx skills are taken over.

• squirrelscan - operating the CLI: install, login, keys, credits, running audits, publishing reports, the entity map, MCP setup, config, troubleshooting.
• audit-website - the full fix loop: audit, map issues to source files, fix in batches, re-audit until the site scores well.

squirrel skills install

Then, in your agent:

Use the audit-website skill to audit this site and fix all issues but only crawl 10 pages

Output formats

Source and development

The complete local CLI, crawler, audit engine, rules, report generators, CLI-facing cloud clients, and documentation site are open source in this repository. The hosted API, website, dashboard, and cloud worker implementations are separate private services.

Prerequisites: Bun 1.4.4 and Git.

git clone https://github.com/squirrelscan/squirrelscan.git
cd squirrelscan
bun install --frozen-lockfile
bun run dev -- audit https://example.com --max-pages 10

Run the same checks used in pull requests:

bun run format:check
bun run lint
bun run typecheck
bun test
bun run build
bun run docs:check
bun run docs:build

See CONTRIBUTING.md before opening a pull request. Contributions require a Developer Certificate of Origin sign-off (git commit -s).

Telemetry

Telemetry is enabled by default and is deliberately minimal: event name, CLI version, a random install ID, and bounded error categories. It does not send API keys, URLs, report contents, or raw error messages. After squirrel auth login, telemetry requests carry your login session (the 90-day sign-in token from login, never an API key) so events tie to your account; signed out, they are anonymous. Nothing is sent when telemetry is off.

Disable it permanently with:

squirrel self settings set telemetry false

Or disable all telemetry and install registration for any invocation by defining NO_TELEMETRY (any defined value works, including an empty value, 0, or false) or DO_NOT_TRACK=1 (or true). A project can opt out with "telemetry": false in .squirrel/settings.json:

NO_TELEMETRY=1 squirrel audit https://example.com

Links

• Website
• Documentation
• FAQ
• Coding agent integration
• Share feedback
• Bugs, issues and feature requests
• Twitter/X

If squirrelscan caught something your agent missed, a star on GitHub helps other developers find it.

License

The CLI and the repository contents are licensed under the MIT License. squirrelscan names and logos are covered by TRADEMARKS.md.