<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
	<channel>
		<title>Citizen Ai on Nick Perkins</title>
		<link>https://nickperkins.au/tags/citizen-ai/</link>
		<description>Recent content in Citizen Ai on Nick Perkins</description>
		<generator>Hugo</generator>
		<language>en</language>
		
		
		
		
			<lastBuildDate>Thu, 15 Oct 2026 10:00:00 +1000</lastBuildDate>
		
			<atom:link href="https://nickperkins.au/tags/citizen-ai/index.xml" rel="self" type="application/rss+xml" />
			<item>
				<title>I Was Wrong About MCPs</title>
				<link>https://nickperkins.au/article/mcp-retraction/</link>
				<pubDate>Thu, 15 Oct 2026 10:00:00 +1000</pubDate>
				<guid>https://nickperkins.au/article/mcp-retraction/</guid>
				<description>&lt;p&gt;In June I wrote about &lt;a href=&#34;https://nickperkins.au/article/skills-over-mcps/&#34;&gt;replacing MCPs with skills and scripts&lt;/a&gt;, and about a skill file beating an MCP server almost everywhere except where there&amp;rsquo;s no CLI. I still work that way. The conclusion I drew from it, that MCP servers are on the way out, was wrong.&lt;/p&gt;&#xA;&lt;h2 id=&#34;the-context-tax-turned-out-to-be-fixable&#34;&gt;The context tax turned out to be fixable&lt;/h2&gt;&#xA;&lt;p&gt;My complaint was concrete. Load an MCP server and every tool definition goes into the context window, whether the agent uses it or not. The Atlassian MCP alone put dozens of schemas in front of the model before I&amp;rsquo;d asked it for anything. That part is still true.&lt;/p&gt;</description>
			</item>
	</channel>
</rss>
