Accessibility findings your coding agent can act on

Accessibility findings usually arrive as a PDF or a dashboard a developer has to re-read and translate by hand. Over MCP they arrive as structured evidence in the tool your agent is already working in — and the same tool can re-check the site once the fix is live.

1. 01

   Assess

   Ask for an assessment of a publicly reachable website, or read an assessment Evolize has already published.

2. 02

   Inspect

   Each finding carries the rule, severity, affected URL, element selector, the relevant markup, the WCAG criterion, and remediation guidance.

3. 03

   Fix in your codebase

   Evidence describes the rendered DOM, not your source tree. Your agent maps it to the right file using its own repository access — Evolize never reads your code.

4. 04

   Verify

   Deploy to a reachable URL, then request a fresh assessment and check whether the issue is still detected on the inspected page. A finding's absence only means a fix if that page appears in the result's inspected-page coverage — an unvisited page reports nothing either.

Add the server once; the three tools then appear to your agent.

claude mcp add --transport http evolize https://evolize.com/api/mcp

{
  "mcpServers": {
    "evolize": {
      "url": "https://evolize.com/api/mcp"
    }
  }
}

https://evolize.com/api/mcp

Replace example.com with the site you want assessed.

Use Evolize to assess the accessibility of example.com. Summarise the observed time and the inspected scope, then list the confirmed violations with their affected URL, element selector and WCAG criterion. Fix the highest-severity issues in this project where the code is in scope, then request a fresh Evolize assessment once the change is deployed to a publicly reachable URL.

Abridged from a real Evolize assessment. The observation time and inspected scope are the actual values.

{
  "rule": "aria-allowed-attr",
  "severity": "high",
  "confirmed_violation": true,
  "wcag": [
    {
      "criterion": "4.1.2",
      "title": "Name, Role, Value",
      "level": "A",
      "relationship": "primary"
    }
  ],
  "user_impact": "Elements must only use supported ARIA attributes",
  "affected_page_count": 4,
  "instance_count": 24,
  "evidence": {
    "affected_url": "https://amano.bz/",
    "selector": ".parent.menu__item[data-main-menu-text-item=\"\"]:nth-child(2)",
    "element_html": "<div class=\"menu__item  parent\" data-main-menu-text-item=\"\" aria-haspopup=\"true\" aria-expanded=\"false\" data-hover-disclosure-toggle=\"dropdown-018713631454ddb8547daf36e704562a\" aria-controls=\"dropdown-018713631454ddb8547daf36e704562a\">",
    "accessible_name": "HOLIDAY\n  \n  \n    \n      \n        \n            \n              \n                Gourmet Snacks & Foods\n              \n            \n              \n                Cocktail Napkins",
    "role": null
  },
  "remediation": "Review the affected element and correct the reported accessibility issue."
}

Two read-only readers and one that can start real work.

get_public_report

Read an assessment Evolize has already published for a website. Read-only: starts no scan. Returns not_found when nothing has been published for that site — which says nothing about whether the site has problems.

scan_website

Request an assessment of a publicly reachable website. Reuses a completed assessment up to 24 hours old; pass refresh=true to require newly executed work after a fix. Returns a handle immediately rather than holding the request open.

get_scan_result

Read the result for a handle, paging through every finding. Read-only and safe to poll: it starts nothing and consumes no scan allowance. While the assessment is running it returns the status and a retry interval.

Free limits

• A completed assessment up to 24 hours old is reused unless you pass refresh=true.
• Up to 3 newly executed assessments per website per 24 hours, refreshes included.
• Evolize's existing per-IP, per-website and platform-wide scan limits also apply; when one is reached the tool returns the reason and a retry time rather than failing opaquely.
• Reading results and polling are free and never consume the scan allowance.

Publication

A qualifying free assessment may be published as a public report at evolize.com/reports/<host> under Evolize's existing publication policy, which is the same policy that applies to an assessment started from the website. Requesting an assessment over MCP does not publish anything by itself, and it does not sign anyone up for email.

What a result does not claim

A completed assessment is not a conformance certification. Automated checks cannot detect every accessibility barrier, and many require human judgement. Results separate confirmed automated violations from findings that need review, and never merge the two.

The assessment and its findings are free. Evolize's paid plans add continuous monitoring, broader coverage, history and team features — so a regression is caught the next time the page changes rather than the next time someone runs a scan.

• How Evolize detects and verifies findings
• Published public reports
• Privacy
• Terms