<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>MCP on Simon Sun</title><link>https://fatflowers.github.io/tags/mcp/</link><description>Recent content in MCP on Simon Sun</description><generator>Hugo -- 0.155.3</generator><language>en</language><lastBuildDate>Wed, 29 Apr 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://fatflowers.github.io/tags/mcp/index.xml" rel="self" type="application/rss+xml"/><item><title>In-Depth Analysis of MiroThinker 1.7: Engineering Optimizations and Guardrails for Long-Horizon Reasoning Agents</title><link>https://fatflowers.github.io/posts/original-tech/miromind-1.7/</link><pubDate>Wed, 29 Apr 2026 00:00:00 +0000</pubDate><guid>https://fatflowers.github.io/posts/original-tech/miromind-1.7/</guid><description>&lt;blockquote&gt;
&lt;p&gt;MiroThinker 1.7 has achieved &lt;a href="https://github.com/MiroMindAI/MiroThinker"&gt;SOTA results&lt;/a&gt; in the field of long-horizon question reasoning. These outstanding results stem from the combination of a powerful Model and a solid Harness. This article documents key engineering optimizations in its Harness implementation.&lt;/p&gt;
&lt;/blockquote&gt;
&lt;h2 id="prerequisite-background"&gt;Prerequisite Background&lt;/h2&gt;
&lt;p&gt;MiroThinker is a deep research Agent — given a complex question (&amp;ldquo;what are the titles of the cs papers on arxiv today&amp;rdquo;), it autonomously breaks down the task, searches, scrapes webpages, runs Python for verification, and finally outputs the &lt;code&gt;\boxed{answer}&lt;/code&gt;. The foundation is classic ReAct: each turn involves LLM reasoning + tool calling, the results are written back to the history, looping 200~300 times until convergence.&lt;/p&gt;</description></item></channel></rss>