Recipes / Make money with AI agents
Made7 paying customersSelf-reported
WhoBuild With MD
SourceX ↗
Date11 Oct 2026

Sell one job to coding agents: a landing-page review they can call over MCP

He spent a year building features nobody paid for. After roasting 500+ landing pages and getting 7 paying customers, he saw what people wanted: someone to tell them why their landing page isn't converting. So he cut FixRoast down to that. You connect it to Cursor or Claude Code over MCP, your agent scans the page, and every issue comes back in one list. The agent fixes them, marks them done, and the next scan checks whether each fix landed.

Setup timeOngoing product
CostYour hosting + model costs
RunsOn demand
NeedsCursor or Claude Code, MCP server
TypeProduct

The template

Adapted from the steps the source described. Fill in the [BRACKETS].

Review the landing page at [URL] (code is in this repo). 1. Take a screenshot and read it like a first-time visitor. What does the page say it does in 5 seconds? 2. List every issue in one list, grouped as: layout/UX, copy, positioning, performance, technical SEO, readiness for AI search and chatbots. 3. For each issue: file:line if it's in the code, the fix, and how sure you are. 4. Skip anything in [SNOOZED LIST]. Propose the fixes as a diff. After I apply them, scan again and mark each issue fixed or still open.

What they did

  • Cut the product to one job after seeing what the 7 paying customers actually wanted.
  • Exposed it over MCP so Cursor or Claude Code can call it from inside the project.
  • One scan returns one list: UI/UX from a model that reads the actual screenshot, copy, positioning, performance, technical SEO and AI readiness.
  • The agent fixes issues and marks them done; the next scan confirms whether each fix landed.
  • Irrelevant issues can be snoozed so they stop coming back. Weekly full-site scans are next on his list.
“500+ pages roasted and 7 paying customers later, the answer was obvious”— Build With MD (@ShipWithMD), X

Summary written by Agent Cookbook. Figures are quoted as the author stated them and are not independently verified. See the original for full context.

Approval rule

🛡
Your agent proposes fixes in your repo; you review the changes before they ship.

Set it up

  1. Add the prompt as a command or skill in your coding agent.
  2. Run it on your own landing page and apply the fixes you agree with.
  3. Re-run to confirm the fixes landed; snooze anything that doesn't apply.
  4. If you sell this as a service, start the way he did: review pages by hand, see what people pay for, then cut to that.

Common failures

⚠
  • Building features before anyone pays; he says a year of that is what led to the cut.
  • Lists so long nobody acts on them; snooze what doesn't matter.

Variations

  • Run it weekly across every page of the site.
  • Offer it as an MCP server other people's agents can call.

Get the next money recipe. Free.

New recipes every week, plus the top 10 agent setups pack.

Unsubscribe anytime. Privacy

More money recipes

All 19 ways people make money with AI agents →

Read original ↗