A search I keep running into: "ExpiredDomains.net MCP". People who already filter domain lists in a browser want to hand the same job to Claude or ChatGPT. Fair enough, but ExpiredDomains.net has no official MCP server and no API. Its FAQ goes further and warns that accounts may be closed when a script, bot or AI agent connects to the member area.

You can find community scrapers for it, and some of them are exposed through MCP. Wrapping a scraper in MCP changes how it gets called. It is still a scraper, it still breaks when the page markup changes, and some of these tools ask for your account credentials. I wouldn't build a client workflow on that.

What a proper expired-domain MCP should let the assistant do

The protocol itself is the easy part. What matters is whether the assistant can run a research process end to end:

  • search current auction, expired, backorder and buy-now inventory
  • apply measurable SEO and history filters instead of guessing what "good" means
  • open a full report when a candidate looks promising
  • save a filter so the same brief runs again next week
  • attach notes and tags to a report
  • create a share link for a teammate or client
  • run a live check right before the buying decision

The Karma.Domains MCP server covers that list with 32 MCP tools on top of a database of 13M+ domains, 400k+ new domains a day from 40+ sources, and 90+ filters. The endpoint is https://mcp.karma.domains/mcp. In ChatGPT or Claude.ai you add the URL and authorize with OAuth in the browser. Clients without OAuth support, like some IDE setups, send an API key as a Bearer token instead. Setup steps for each client are on the Karma.Domains MCP page.

How the conversation goes in practice

Start with a buying brief rather than "find good domains":

> Find expired .com and .org domains related to home improvement. Require at least 20 referring domains and three years of Wayback history. Exclude gambling, pharma, adult content and obvious redirect histories.

The assistant turns that into filters and returns a shortlist with report links. The useful part comes next, when you refine in plain language: keep the exclusions, require Moz DA of at least 20, show only names under 18 characters, and widen to .net if fewer than five match. In a filter panel you would rebuild half the query for that.

Once a brief works, save it under a name like client-a-home-improvement and rerun it next week. For an agency this matters more than the chat itself, because everyone runs the same criteria and changes to the profile are visible.

For candidates that survive, write the reasoning into the report: tags for filtering later, and a note that says what passed and what still needs checking. "Looks good" tells the next reviewer nothing.

Live checks (availability, authority, spam score) belong at the very end, on one or two finalists, because those values change and the checks cost credits.

Where the assistant stops

The assistant can narrow thousands of records to a handful and explain why each one is there. It should not place bids. A person still reads the archive, looks at the marketplace listing, thinks about trademark risk and makes the call.