Agent-Ready Web: What MCP Means for Accessibility Tools

In November 2024, Anthropic published the specification for the Model Context Protocol (MCP) — a standard that lets AI agents interact with external tools and services in a structured, consistent way. MCP is to AI agents what REST was to web services: a common interface that lets systems talk to each other without custom integration work for every combination.

Accessibility tools are a particularly natural fit for MCP. Think about what an accessibility tool does: it scans a page, identifies rule violations, explains each violation in plain language, suggests a fix, and tracks compliance status over time. Every one of those capabilities is something an AI agent would benefit from having access to — and MCP is exactly the mechanism for providing that access.

A Quick MCP Primer

MCP works through a client-server model. An MCP server exposes a set of “tools” — functions that can be called with structured inputs and that return structured outputs. An MCP client (like Claude, or any agent that implements the protocol) discovers which tools are available, calls them when relevant, and uses the results in its reasoning.

The protocol supports three transport modes: stdio (for local tools running as subprocesses), SSE (server-sent events for streaming), and HTTP (for hosted services accessible over the network). Certvo uses HTTP transport, which means you can call the Certvo accessibility API from any MCP-compatible agent without running anything locally.

MCP also defines “resources” (static data the server exposes, like a documentation page or a compliance report) and “prompts” (reusable prompt templates the server suggests). The tools primitive is the most important for accessibility use cases.

Why Accessibility Fits MCP Perfectly

An agent working on a codebase can now do things that were previously impossible without custom tooling:

• Scan during development: While fixing a bug, an agent can scan the affected page to check whether the fix introduced any accessibility regressions — without leaving the coding workflow.
• Fix in context: When a scan returns a violation, the agent can request the AI-generated fix, understand it, and apply it as part of a larger change — with full context about what else it is changing and why.
• Explain to non-technical stakeholders: An agent can scan a page, get the results, and write a plain-language summary for a product manager — without the PM ever touching a JSON report.
• Enforce in CI: A CI agent can scan every pull request and comment on the PR with violations and suggested fixes, without any custom GitHub Actions configuration.

Adding Certvo to Claude: The Real Command

Certvo's MCP server is live and accessible at https://certvo.com/api/mcp/. To add it to Claude:

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

That is the complete command. After running it, Claude will list Certvo's tools when you ask “what tools do you have?” and will call them automatically when you ask accessibility-related questions. You can verify the connection with:

claude mcp list
# Output:
# certvo  https://certvo.com/api/mcp/  (http)

A Live Walkthrough: From Prompt to Fix

Here is what an actual session looks like after adding the Certvo MCP server. These are the exact prompts and tool calls from the /docs/mcp documentation.

You type:

Scan https://example-shop.com/checkout for accessibility issues and
summarize the critical violations with suggested fixes.

Claude calls the Certvo MCP tool:

{
  "tool": "certvo_scan",
  "arguments": {
    "url": "https://example-shop.com/checkout",
    "wcag_level": "AA",
    "include_fixes": true
  }
}

The Certvo server scans the page and returns:

{
  "scan_id": "sc_2026050413xy",
  "url": "https://example-shop.com/checkout",
  "score": 62,
  "violations": [
    {
      "rule": "label",
      "impact": "critical",
      "wcag": "3.3.2",
      "count": 4,
      "fix": "<label for='card-number'>Card number</label>\n<input type='text' id='card-number' ...>",
      "confidence": 94
    },
    {
      "rule": "color-contrast",
      "impact": "serious",
      "wcag": "1.4.3",
      "count": 7,
      "fix": ".checkout-label { color: #374151; } /* was #9ca3af */",
      "confidence": 88
    }
  ],
  "report_url": "https://certvo.com/report/sc_2026050413xy"
}

Claude uses this to produce a response that explains the issues in plain language, shows the fixes, and links to the full report. You can follow up:

Apply the label fixes to the checkout form component and show me the diff.

Claude now has both the scan results (from the MCP tool) and your codebase (from the file system or a repository tool). It can generate the complete diff applying the fixes — something that was previously impossible without custom integration.

The 14 Agent-Ready Checks

Certvo's /agent-ready page lists 14 protocols and standards that the platform checks for agent compatibility — not just accessibility, but also structured data, API versiveness, and the technical properties that make a site reliably scrapable and interactive for AI agents. The full list is on that page, but the key idea is that “agent-ready” and “accessible” are more correlated than most people realize: both require semantic HTML, predictable structure, and keyboard-operable interactions.

Beyond Scanning: What Else MCP Enables

The Certvo MCP server exposes more than just scanning. From the MCP documentation:

• certvo_get_report: Retrieve a previous scan report by ID — useful for tracking changes over time
• certvo_check_wcag: Check a specific HTML snippet against a single WCAG success criterion — useful when an agent is unsure whether a particular pattern is compliant
• certvo_generate_vpat: Trigger VPAT generation for a scanned product
• certvo_explain: Get a plain-language explanation of a WCAG criterion — useful when an agent needs to communicate the “why” of an accessibility requirement to a non-technical stakeholder

Getting Started

If you use Claude regularly and work on web projects, the single most impactful thing you can do right now is add the Certvo MCP server with the one-line command above. It costs nothing — the MCP endpoint is available to all Certvo users, including free tier — and it means that from now on, whenever you ask Claude anything about a web page, it can check accessibility without you having to think about it.

For teams wanting to integrate the MCP endpoint into their own agents and CI workflows, the full API documentation is at /docs/mcp. Or start with a free scan to see the kind of output your agent will be working with.