Urgent.News

What's breaking now, across thousands of outlets.

AI

One MCP server, two rulebooks: shipping to Claude and ChatGPT

Sitecomb runs automated website quality audits. You paste a web address and it looks at six areas: security, privacy and legal, page health, works on phones, contact details, and found online. We also run it as a hosted MCP server, so you can ask Claude "is my website any good?" and get a real check back, not a guess. Here are five things we learned building that server. None of them are clever.…

Sitecomb is an automated website quality auditing service that can be run as a hosted MCP server, enabling users to ask Claude about the quality of their website. In this story, five lessons learned from building the server are shared. The first lesson is that the server runs statelessly, with a fresh server created for every request.

While this may seem wasteful, it doesn't require storing sessions, expiring data, or managing tool calls. The maximum runtime for a check is limited to 120 seconds, which includes the time to load the site and perform the check. This is crucial as the actual workload can take up to a minute. The second lesson is that the server has two host policies due to differing rules from OpenAI and Claude.

OpenAI's app rules prohibit upselling a digital product or showing prices, while Claude's directory allows linking to a paid report. Therefore, the server has one definition mounted twice with a slight policy object. The third lesson is that the model is instructed not to invent problems, fixes, or quality scores. Instead, it should report findings as they are and never add additional issues.

The fourth lesson is that public tools need limits, with each check costing real compute. The limits are set cheaply, starting with bad input and counters before the expensive browser load. The fifth and final lesson is that the tool description should be tailored to the questions people actually ask. Instead of using generic terms, the tool description should directly address common queries about website quality and provide clear guidance on its usage and limitations.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

This story

This is one outlet's version. Read the fullest account.

Every outlet that reported this, and when →

Read the original at dev.to →

More in AI

How I Built AI-Native Products as a One-Person Software Factory

I didn't use AI agents to skip product discipline. I used them to compress it. The result was a one-person software factory with the same stages I expect in a serious v1.

  • Author used AI to accelerate various software development stages, not to bypass them.
  • Released open-source AI and agentic projects like AgentEval Forge and ToolTrust.

Building a family budget bot on $0: Telegram, free-tier AI fallbacks, and Arabic RTL

For four years I tracked my family's spending in an Excel sheet. It died the way most budget sheets die: logging a purchase took too long, and nobody but me ever opened the file.

  • Mizan created a Telegram bot to help families track spending using free-tier services
  • Bot architecture includes Next.js, Hono, Supabase, Drizzle, and Netlify free hosting

SpecGuard: When AI Coding Agents Write Tests That Approve Bugs

Ask a coding agent to write checks for a requirement and it will often copy what the existing code does instead. In SpecGuardBench, the best agent wrote a faithful validation harness for only 63% of…

  • SpecGuardBench finds only 63% of coding agents produce faithful harnesses from requirements.
  • Buggy reference code reduces agents' faithfulness to requirements by 19 to 31 points.
  • SpecGuard tests users' ability to distinguish correct, buggy, and valid alternative implementations.

More from Sunday 15 November →