For developers and platform builders

Find a problem. Fix it. Check again.

Bring source evidence into your development workflow. Inspect a public page, save the result, and compare its reported findings after your next change.

Run your first check Download developer tools

Node.js 22.13+ · No SDK dependencies · Public-source evidence · Experimental WebMCP

A repeatable check from your terminal

Download and extract the toolkit, then replace the example URL with your public page. No repository access, dependency installation, or application deployment is needed. Use a new output filename for each scan.

# Download and extract the developer tools ZIP, then open that folder.
node integrations/developer-kit/bin/iswebmcp.mjs scan https://example.com --output baseline.json
# After changing your public page, save a second result:
node integrations/developer-kit/bin/iswebmcp.mjs scan https://example.com --output current.json
node integrations/developer-kit/bin/iswebmcp.mjs compare baseline.json current.json

Queries are removed before transmission by default. The hosted scanner cannot reach localhost, private networks, or signed-in pages. Saved JSON stays on your machine; the service may retain a sanitized URL attempt for 90 days as described in our privacy policy.

Read the data policy

Inside your coding agent

Check → implement → compare

Connect a Streamable HTTP MCP client to https://iswebmcp.com/mcp. Ask for a source audit, request a matching implementation recipe, review the code locally, then compare your saved before-and-after summaries.

“Audit my public search page. Show me the accessible-controls recipe. Explain the proposed changes before editing. After I deploy, compare my baseline and current source summaries. Keep runtime claims separate.”

Six tools on one endpoint, including get_implementation_recipe and compare_source_reports. Your agent needs separate local editing permission; this server cannot change your repository, deploy, or control your browser. Only send summaries you are comfortable sharing with the service and your agent host. Use the local CLI for sensitive evidence.

Explore implementation recipes

Embed evidence in your platform

Use the Node SDK from your backend or connect an MCP client. Keep findings, recommendations, collection coverage, and missing evidence together in your own interface.

Download SDK with examples

Check changes in GitHub Actions

Run a scan against your public deployment and compare a reviewed baseline. Incomplete or incompatible evidence produces an inconclusive check, so missing data cannot silently pass.

Download CI adapter and setup guide

What this check tells you

Version 0.2 compares the complete finding inventory returned for each bounded source scan, using stable rule identifiers and matching scan-input fingerprints. A passing comparison means it found no new or worsened failing findings in comparable summaries. Existing issues may remain. It does not execute your tools, verify an authenticated journey, or measure agent success.

Upgrading from 0.1? Capture a fresh baseline with the new toolkit. Older summaries lack the provenance required for release comparisons. Model changes also require a new, reviewed baseline; do not treat scores from different model versions as improvements or regressions.

Runtime verification needs a connected browser, a defined task, and an observed outcome. A general browser adapter and durable hosted monitoring are planned; the current toolkit saves evidence locally and uses the shared, rate-limited public scanner.

Understand the evidence

Design partner pilots

Bring one workflow you want to improve.

We are looking for developers building agent interactions and platforms that want to embed useful audit evidence. Start with a public test page, one recurring problem, and a concrete way to verify the fix.

Discuss a pilot