<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Recipes on pgmi</title><link>https://vvka-141.github.io/pgmi/docs/recipes/</link><description>Recent content in Recipes on pgmi</description><generator>Hugo</generator><language>en</language><atom:link href="https://vvka-141.github.io/pgmi/docs/recipes/index.xml" rel="self" type="application/rss+xml"/><item><title>Semantic MCP curation</title><link>https://vvka-141.github.io/pgmi/docs/recipes/semantic-mcp-tool-curation/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://vvka-141.github.io/pgmi/docs/recipes/semantic-mcp-tool-curation/</guid><description>&lt;h1 id="recipe-semantic-agent-tool-curation"&gt;Recipe: Semantic Agent-Tool Curation&lt;a class="anchor" href="#recipe-semantic-agent-tool-curation"&gt;#&lt;/a&gt;&lt;/h1&gt;
&lt;blockquote class='book-hint '&gt;
&lt;p&gt;&lt;strong&gt;Use this only at tool-overload scale.&lt;/strong&gt; When an agent has &lt;em&gt;dozens&lt;/em&gt; of tools,
exposing all of them at once dilutes the model&amp;rsquo;s attention and inflates token
cost. This recipe surfaces the relevant subset for a given query. If your
project has a handful of tools, skip it — expose handlers directly through the
existing MCP/RPC routing and let the client list them all.&lt;/p&gt;
&lt;/blockquote&gt;&lt;p&gt;This recipe is a &lt;strong&gt;cookbook pattern&lt;/strong&gt;, not part of the advanced template. It
adds an embedding dependency (pgvector + an external embedding model) that
narrows managed-cloud compatibility and adds runtime cost, so it is deliberately
kept out of &lt;code&gt;internal/scaffold/templates/&lt;/code&gt;. Build it in your own project on top
of the handler registry the advanced template already ships.&lt;/p&gt;</description></item></channel></rss>