MCP Agent Accessibility Auditor

What the mcp-agent-accessibility-auditor MCP server does

The mcp-agent-accessibility-auditor MCP server provides an MCP wrapper around Mamba Labs' Agent Accessibility Auditor actor on Apify. It accepts one company domain per call and returns the actor's result without transforming it. The result is a flat record with 42 fields grouped around several kinds of machine-readable access:

• llms.txt availability and validation
• Robots AI crawler rules, including Content Signal directives
• Structured data presence and health
• Render mode
• Machine-readable endpoint discovery

The service is intended for technical SEO or growth work where a developer needs concrete evidence about how a site may be read by AI crawlers and agents.

How it works

The MCP tool starts an Apify actor run, waits for it to finish, and returns the resulting dataset. It does not call a language model while auditing, so repeated results reflect fetched site data rather than model interpretation. The actor checks robots.txt first, then the homepage and up to ten probes concurrently. A typical run takes about two to four seconds.

The audit can use a seven-day cache. Setting skipCache to true requests a new run; otherwise, a repeat audit within the cache period does not create a new domain charge. The actor run is allowed up to 1,800 seconds. If it remains active two minutes after that limit, the tool stops waiting and returns the run ID and an Apify Console link.

Setup and configuration

Install the package with:

npx -y @mambalabsdev/mcp-agent-accessibility-auditor

Set APIFY_TOKEN in the MCP client's environment. The token is obtained from the Apify account integrations page. A Claude Desktop configuration uses npx as the command, passes the package name with -y, and supplies the token through the server's env object.

Tools and capabilities

The mcp-agent-accessibility-auditor MCP server exposes one tool: audit_agent_accessibility.

Its required domain input accepts a domain such as vercel.com; full URLs are also accepted, with protocol and path removed. Each call handles one domain. Optional controls include:

• checks: select checks such as llms_txt, robots_ai, sitemap, openapi, security_txt, feeds, json_ld, microdata, open_graph, canonical, and render_mode
• check_endpoints: include or omit sitemap, OpenAPI, security.txt, and feed checks
• check_structured_data: include or omit JSON-LD, microdata, Open Graph, and canonical checks
• skipCache: bypass the seven-day cache

Omitted checks return null, and the score is recalculated using only the selected checks. The result can identify why a successful /llms.txt response was not accepted, including cases where the body was not real Markdown.

Limitations and notes

This is a thin client for the Apify actor; the actor supplies the audit behavior and output fields. Running it consumes Apify credits and charges per analyzed domain, along with a small actor-start fee. Errors such as invalid input, an invalid or exhausted token, timeouts, unsuccessful actor runs, or non-dataset results are returned explicitly rather than as empty results.

The tool reports fetched facts and does not determine accessibility through model judgment. It also does not silently recover failed runs. If a long run exceeds the waiting behavior, use the returned run ID and Apify Console link to inspect the eventual result.