Last updated: November 20, 2026
Once the free WebAbility MCP server is connected to your editor, you need to know what to ask it. This post gives six prompts for jobs that come up every week, what the assistant does with each one, and where a person still has to decide.
Before you start
Point your editor at the hosted server, https://mcp.webability.io/mcp. It needs no account and no API key. The MCP page has the one-line setup for Claude Code, Cursor, GitHub Copilot in VS Code, Windsurf and Claude Desktop, the full tool list, a local install, and a way to scan your localhost dev server.
1. Check a component before you open a pull request
Scan this component's HTML with scan_html and list the issues by WCAG criterion. Fix the ones with a known value.
scan_html runs without a browser, so it is fast enough to use on every change. It checks structure: labels, accessible names, roles and ARIA. It does not render anything, so contrast waits for prompt 3.
2. Scan a page and fix what is certain
Scan http://localhost:3000/pricing with scan_page. Fix the mechanical issues in the source, then run verify_fix on each one.
Every issue comes back with a fix type. A mechanical fix has a known value, such as a missing attribute, and the assistant can apply it as given. A contextual fix, such as alt text or a form label, needs the assistant to read the markup and write the value. Read what it wrote before you commit.
3. Check contrast on the rendered page
Check the contrast of the primary button on /pricing with check_color_contrast and suggest a passing color close to our brand color.
Contrast depends on what the browser draws, so the tool proposes a value and does not apply it. A brand color change is a design decision; get a designer to approve it.
4. Catch regressions before a release
Run diff_scan on /checkout against the scan from before this change and show only the new issues.
diff_scan splits the result into fixed, new and remaining, which keeps a release check short enough to read.
5. Scan a whole journey as one report
Use flow_scan with autoNavigate on /, /pricing and /signup and give me one report.
flow_scan takes a list of URLs and merges the results. It does not click buttons or log in. For pages behind a login, use the local install with a saved browser session; the MCP page shows how.
6. Ask what a rule checks
Use get_rules to explain what the link purpose rule (WCAG 2.4.4) checks.
This helps when a reviewer asks why a fix was needed, and it keeps the answer tied to the rule the scanner ran.
Where a person still decides
The MCP page puts automated testing at roughly 30 to 40 percent of WCAG issues. A scan can see that an image has alt text but not whether the text is useful. It can see the focus order but not whether that order makes sense to someone using a keyboard or a screen reader. Those checks need a person, and some need a full accessibility audit.
Automation fixes the repeatable issues on the live page, and a human audit covers the rest. Neither alone is compliance. The MCP server handles the first part while you build; the Accessibility Enhancer handles it on the live site.
Start here
Open the WebAbility MCP page, copy the line for your editor, and run prompt 2 on a page you are working on today.






