DEV Community

carsonroell-debug
carsonroell-debug

Posted on

I Built an MCP Server That Finds Broken Links. Here's How AI Agents Use It

#ai #mcp #agents #showdev

I published an MCP server that lets Claude, Cursor, and any MCP-compatible AI agent scan websites for broken links, estimate revenue impact, and suggest fixes. Here's why I built it and how it works.

Every day, my outreach engine scans affiliate blogs and finds the same thing: sites with 100+ pages almost always have 20-30 broken affiliate links that nobody knows about. Each one is a leak in the revenue bucket.

The auditing part was automated. But acting on the results still required me to look at the report, figure out what to fix, and make the changes.

That's exactly what AI agents are good at.

The Model Context Protocol lets AI agents call tools through a standard interface. Instead of giving an agent a screenshot or a CSV and saying "figure this out," you give it structured tools that return structured data.

So I wrapped LinkRescue's scanning engine as an MCP server. Now any agent can:

  1. Scan a site → get back every broken link with status codes, page locations, and revenue estimates
  2. Get fix suggestions → receive prioritized remediation steps with code snippets
  3. Set up monitoring → schedule recurring checks

The core tool. Give it a URL, it crawls the site and returns a structured report.

An agent in Cursor could take this output and directly edit the source files. An agent in Claude could draft the update instructions for a team.

Set up scheduled monitoring and verify connectivity. Simple but necessary for agent workflows that run unattended

The server falls back to realistic simulated data when no backend API is connected, so you can test immediately without any setup.

Why MCP > API for This

A REST API returns data. An MCP server returns data that agents understand natively. The difference:

  • REST API: Agent calls endpoint → parses JSON → figures out what to do → makes another call
  • MCP server: Agent discovers tools → reads descriptions → chains them together automatically

With MCP, I've watched Claude scan a site, get the broken links report, pipe it directly to get_fix_suggestions, and present a prioritized fix plan — without any prompting about the workflow. The tool descriptions guide the agent.ive gone

The MCP server is open source. The full SaaS at linkrescue.io has additional features (dashboard, email alerts, historical tracking, team management).

I'm curious what other "scan and fix" workflows would benefit from MCP servers. If you've built something similar, I'd love to see it.

Links:

Top comments (0)