The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the WSG Check listing page.
Check a website against automated heuristics based on the W3C Web Sustainability Guidelines. Get a weighted score, recommendations, and reports in terminal, JSON, Markdown, or HTML format through a CLI or a Next.js web app.
The latest release is v0.6.0.
It targets WSG July-2026 (Group Note Draft). It includes an MCP server for AI assistants, with the check_url, list_guidelines, and get_guideline tools, listed in the MCP Registry as io.github.ivanoats/wsg-check. The server first shipped in v0.4.0; v0.3.0 was tagged on GitHub but not published to npm. See the changelog and spec versioning policy.
Scores from different spec versions should not be treated as directly comparable. The package version and check selection also matter when comparing reports.
Requires Node.js 22 or later. Pin the released CLI for reproducible usage:
To run the current source, including unreleased features, use Node.js 22.22.1 or later (required by development tooling):
See CONTRIBUTING.md for environment configuration and development commands.
false.WSG-Check ships with a command-line tool that lets you check any website directly from your terminal or integrate checks into CI pipelines.
| Option | Alias | Description | Default |
|---|---|---|---|
--format <format> | -f | Output format: terminal, json, markdown, html | terminal |
--output <path> | -o | Write report to a file instead of stdout | (stdout) |
--categories <list> | -c | Comma-separated categories: ux,web-dev,hosting (business planned — no automated checks yet) | all |
--guidelines <list> | -g | Comma-separated guideline slugs to run, e.g. minify-and-remove-unused-code (numeric IDs such as 3.3 are deprecated) | all |
--fail-threshold <n> | Exit code 1 if overall score < n (0–100) | 0 | |
--verbose | -v | Enable verbose logging | false |
--config <path> | Path to wsg-check.config.json or .wsgcheckrc.json | (auto-discover) | |
--version | Print version and exit | ||
--help | Print help and exit |
Use --fail-threshold to fail your pipeline when a site's sustainability score drops:
| Code | Meaning |
|---|---|
0 | Check completed and score is at or above --fail-threshold |
1 | Fetch/parse error, or score is below --fail-threshold |
The package includes wsg-check-mcp, a Model Context Protocol server for Claude Code, Claude Desktop, VS Code, Cursor, and other MCP clients. It runs on your machine over stdio, with no account, port, or hosted service. Your assistant can then check a deployed site or your local dev server, look up guidelines, and fix what it finds.
Claude Desktop (claude_desktop_config.json) and Cursor (.cursor/mcp.json):
npx -y @sustainablewebsites/wsg-check --mcp starts the same server; it's the command the MCP Registry listing, io.github.ivanoats/wsg-check, uses.
VS Code (.vscode/mcp.json) uses a servers key:
Try prompts such as "Check http://localhost:3000 against the WSG and fix the top three issues" or "Which WSG guidelines cover web fonts?"
| Tool | What it does |
|---|---|
check_url | Checks a page and returns the score, grade, and each failed or warned check with a recommended fix. detail: "full" adds the complete report. |
list_guidelines | Lists the WSG guidelines, filtered by category, testability, or text, with how many automated checks implement each one. |
get_guideline | Returns one guideline's description and W3C specification link. |
http://localhost:3000 and other loopback addresses work without setup. The trade-off: text the assistant reads, such as a web page or a file in your project, could try to steer it into requesting a service on your machine. Requests are GET-only and the results go back to the assistant. If you don't check a local dev server, add --no-local after wsg-check-mcp in the arguments, or set WSG_CHECK_NO_LOCAL=1.--allow-private-network or set WSG_CHECK_ALLOW_PRIVATE=1.check_url makes is a green-hosting lookup: it sends the hostname to the Green Web Foundation. Local hostnames are never sent.The repository also ships an Agent Skill, wsg-sustainability-review. It walks an assistant through checking a site, fixing the issues in your codebase in priority order, and checking again. It uses the MCP tools when they are configured and falls back to the CLI otherwise. Install it into Claude Code, Cursor, Copilot, and other agents with:
The web app exposes Next.js Route Handlers:
| Endpoint | Purpose |
|---|---|
POST /api/check | Run a check and return the completed report and ID |
GET /api/check/:id | Retrieve a report from the process-local result store |
GET /api/guidelines | List guidelines; current source includes pinned spec provenance in spec |
GET /api/guidelines/:id | Look up a guideline; current source accepts slugs and legacy numeric aliases |
GET /api/health | Health status; current source includes package version and specVersion |
GET /api/openapi | OpenAPI 3.1 JSON document (no interactive documentation UI) |
Responses use CORS headers, shared error envelopes, and in-memory rate limiting. Stored results expire after one hour and the store holds at most 500 entries. Results and limits are not shared across server instances. The browser carries newly created reports through sessionStorage; result URLs are not durable share links. See runtime and storage.
The CLI, web API, and MCP server share a TypeScript fetch → check → score → report pipeline. The code is layered and framework-independent at its core, with direct dependencies on concrete utility implementations. The architecture overview documents these boundaries and the external-I/O exceptions.
Apache-2.0. See NOTICE and the Contributor License Agreement.