Skip to content

an open-source alternative to the dots, bots and muses of the world, run an office of claude code/codex like agents on your laptop, sandboxes or anywhere, uses your existing subscriptions

28.1 KB
Raw Blame
<!doctype html>
<html lang="en">
<head>
<!-- Google tag (gtag.js), Google Ads AW-18388216863. Loaded only when Do Not Track is
     off, so it follows the same rule as the PostHog block on this site. -->
<script>
  window.dataLayer = window.dataLayer || [];
  function gtag(){ dataLayer.push(arguments); }
  if (navigator.doNotTrack !== '1') {
    var gtagScript = document.createElement('script');
    gtagScript.async = true;
    gtagScript.src = 'https://www.googletagmanager.com/gtag/js?id=AW-18388216863';
    document.head.appendChild(gtagScript);
    gtag('js', new Date());
    gtag('config', 'AW-18388216863');
  }
</script>

<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>MCP Tool Poisoning: How to Secure the Model Context Protocol — Munder Difflin Blog</title>
<meta name="description" content="How the Model Context Protocol gets attacked — tool poisoning, rug pulls, command injection — and the layered defenses that secure your MCP tools." />
<link rel="canonical" href="https://munderdiffl.in/blog/mcp-security-tool-poisoning/" />

<link rel="icon" type="image/png" href="https://munderdiffl.in/logo.png" />
<link rel="apple-touch-icon" href="https://munderdiffl.in/logo.png" />
<meta name="theme-color" content="#F5F2E8" />

<!-- Open Graph -->
<meta property="og:site_name" content="Munder Difflin Blog" />
<meta property="og:type" content="article" />
<meta property="og:title" content="MCP Tool Poisoning: How to Secure the Model Context Protocol — Munder Difflin Blog" />
<meta property="og:description" content="How the Model Context Protocol gets attacked — tool poisoning, rug pulls, command injection — and the layered defenses that secure your MCP tools." />
<meta property="og:url" content="https://munderdiffl.in/blog/mcp-security-tool-poisoning/" />
<meta property="og:image" content="https://munderdiffl.in/blog/assets/media/mcp-security-tool-poisoning/hero.png" />
<meta property="og:image:width" content="1600" />
<meta property="og:image:height" content="900" />
<meta property="og:image:alt" content="Illustration: MCP Tool Poisoning: How to Secure the Model Context Protocol" />
<meta property="og:locale" content="en_US" />
<meta property="article:published_time" content="2026-06-05T00:00:00.000Z" />
<meta property="article:author" content="Chaitanya Giri" />
<meta property="article:tag" content="Security" />
<meta property="article:tag" content="MCP" />
<meta property="article:tag" content="Protocols" />
<meta property="article:tag" content="Guardrails" />


<!-- Twitter -->
<meta name="twitter:card" content="summary_large_image" />
<meta name="twitter:title" content="MCP Tool Poisoning: How to Secure the Model Context Protocol — Munder Difflin Blog" />
<meta name="twitter:description" content="How the Model Context Protocol gets attacked — tool poisoning, rug pulls, command injection — and the layered defenses that secure your MCP tools." />
<meta name="twitter:image" content="https://munderdiffl.in/blog/assets/media/mcp-security-tool-poisoning/hero.png" />
<meta name="twitter:image:alt" content="Illustration: MCP Tool Poisoning: How to Secure the Model Context Protocol" />

<link rel="alternate" type="application/atom+xml" title="Munder Difflin Blog" href="https://munderdiffl.in/blog/feed.xml" />
<link rel="preconnect" href="https://fonts.googleapis.com" />
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin />
<link href="https://fonts.googleapis.com/css2?family=Fraunces:opsz,[email protected],500;9..144,600;9..144,700;9..144,900&amp;family=Source+Serif+4:opsz,[email protected],400;8..60,600;8..60,700&amp;family=Space+Grotesk:wght@500;600;700&amp;family=JetBrains+Mono:wght@400;600&amp;display=swap" rel="stylesheet" />
<link rel="stylesheet" href="/blog/assets/blog-press.css" />

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "BlogPosting",
  "headline": "MCP Tool Poisoning: How to Secure the Model Context Protocol",
  "description": "How the Model Context Protocol gets attacked — tool poisoning, rug pulls, command injection — and the layered defenses that secure your MCP tools.",
  "image": ["https://munderdiffl.in/blog/assets/media/mcp-security-tool-poisoning/hero.png"],
  "datePublished": "2026-06-05T00:00:00.000Z",
  "dateModified": "2026-06-05T00:00:00.000Z",
  "author": { "@type": "Person", "name": "Chaitanya Giri" },
  "publisher": {
    "@type": "Organization",
    "name": "Munder Difflin",
    "logo": { "@type": "ImageObject", "url": "https://munderdiffl.in/logo.png" }
  },
  "mainEntityOfPage": { "@type": "WebPage", "@id": "https://munderdiffl.in/blog/mcp-security-tool-poisoning/" },
  "keywords": "mcp tool poisoning, mcp security, model context protocol security, mcp prompt injection, securing mcp servers",
  "articleSection": "Guides",
  "inLanguage": "en-US",
  "url": "https://munderdiffl.in/blog/mcp-security-tool-poisoning/"
}
</script>
<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "BreadcrumbList",
  "itemListElement": [
    { "@type": "ListItem", "position": 1, "name": "Home", "item": "https://munderdiffl.in/" },
    { "@type": "ListItem", "position": 2, "name": "Blog", "item": "https://munderdiffl.in/blog/" },
    { "@type": "ListItem", "position": 3, "name": "MCP Tool Poisoning: How to Secure the Model Context Protocol", "item": "https://munderdiffl.in/blog/mcp-security-tool-poisoning/" }
  ]
}
</script>

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    { "@type": "Question", "name": "What is MCP tool poisoning?", "acceptedAnswer": { "@type": "Answer", "text": "Tool poisoning is an attack where malicious instructions are hidden in an MCP tool's metadata — its name, description, or schema. Because an MCP client passes those descriptions to the model as input the user never sees, a poisoned description can steer the agent into unsafe actions. It's a form of indirect prompt injection that exploits the client-server trust model." } },
    { "@type": "Question", "name": "What is an MCP rug pull attack?", "acceptedAnswer": { "@type": "Answer", "text": "A rug pull is a time-delayed tool poisoning. A server looks benign when you approve it, then later updates its tool descriptions to add malicious instructions — which the agent will follow on the next call. The defense is to treat tool descriptions as untrusted: scan them at install and on every update, and reject silent changes." } },
    { "@type": "Question", "name": "What's the most common MCP security risk?", "acceptedAnswer": { "@type": "Answer", "text": "Despite the attention tool poisoning gets, industry analyses point to something more mundane: token mismanagement — a leaked or over-privileged credential — as the most common real-world MCP breach vector. Least privilege and secret hygiene matter more than defending any single clever injection chain." } }
    
  ]
}
</script>


<script>
  // Anonymous blog analytics (PostHog) — same posture as the main site and the
  // app's TELEMETRY.md: public write-only token ('' = inert), anonymous-only
  // (person_profiles 'never'), no autocapture/recording, DNT respected.
  (function () {
    var PH_KEY = 'phc_qJzsokwcbtTGHemnvywR3nSqivNNgMjHeNVEWkZNydFB';
    if (!PH_KEY || navigator.doNotTrack === '1') return;
    var s = document.createElement('script');
    s.async = true;
    s.src = 'https://us-assets.i.posthog.com/static/array.js';
    s.onload = function () {
      if (!window.posthog) return;
      window.posthog.init(PH_KEY, {
        api_host: 'https://us.i.posthog.com',
        person_profiles: 'never',
        autocapture: false,
        before_send: function (event) {
          // PostHog fills $ip from the connection unless the event carries one.
          // Null it here and disable the GeoIP lookup so no IP, city, postal
          // code or lat/long is ever derived or stored for this page view.
          if (event && event.properties) {
            event.properties.$ip = null;
            event.properties.$geoip_disable = true;
          }
          return event;
        },
        disable_session_recording: true,
        respect_dnt: true
      });
    };
    document.head.appendChild(s);
  })();
</script>
</head>
<body class="theme-press">
<a class="skip-link" href="#main">Skip to content</a>
<nav id="nav">
  <a class="brand" href="https://munderdiffl.in" aria-label="Munder Difflin home">
    <img class="brand-logo" src="https://munderdiffl.in/logo.png" width="1318" height="840" alt="Munder Difflin, Inc. — Multi-Agent Harness" />
  </a>
  <span class="spacer"></span>
  <div class="links">
    <a href="/blog/">Latest</a>
    <a href="/blog/topics/">Topics</a>
    <a href="/blog/about/">About</a>
    <a href="https://munderdiffl.in">↗ Product</a>
  </div>
  <div class="nav-actions">
    <a class="btn sm" href="https://munderdiffl.in/blog/feed.xml" aria-label="RSS feed">⤵ RSS</a>
    <a class="btn sm primary" href="https://munderdiffl.in/#install">⤓ Download</a>
  </div>
</nav>

<main id="main">

<article class="t-guides">
  <header class="post-head wrap">
    <nav class="crumbs" aria-label="Breadcrumb">
      <a href="/blog/">Blog</a>
      <span>/</span>
      <a href="/blog/topics/guides/">Guides</a>
    </nav>
    <h1>MCP Tool Poisoning: How to Secure the Model Context Protocol</h1>
    <p class="dek">How the Model Context Protocol gets attacked — tool poisoning, rug pulls, command injection — and the layered defenses that secure your MCP tools.</p>
    <div class="byline">
      <span class="av" aria-hidden="true">CG</span>
      <strong>Chaitanya Giri</strong>
      <span class="dot" aria-hidden="true">·</span>
      <time datetime="2026-06-05">Jun 5, 2026</time>
      <span class="dot" aria-hidden="true">·</span>
      <span>5 min read</span>
      <span class="tag kind">Guides</span>
    </div>
  </header>

  <div class="hero-wrap wrap"><figure class="fig fig-hero"><img src="/blog/assets/media/mcp-security-tool-poisoning/hero.png" alt="Illustration: MCP Tool Poisoning: How to Secure the Model Context Protocol" loading="lazy" decoding="async" /></figure></div>

  <div class="post-wrap wrap">
    <div class="prose">
      <div class="callout tldr"><span class="ic">TL;DR</span><p>MCP's power — letting an agent discover and call
external tools — is also its attack surface. The core flaw is a <strong>trust model</strong>: an MCP
client hands the model tool <em>descriptions</em> it got from a server, usually <strong>without validating
them</strong>. That makes tool metadata an injection vector (<strong>tool poisoning</strong>), enables
time-bomb <strong>rug pulls</strong>, and — with unsanitized arguments — real command execution. Yet the
most common real breach is mundane: a <strong>leaked or over-privileged token</strong>. Defense is
layered: treat tool metadata as untrusted, scan and pin it, sanitize arguments, enforce OAuth 2.1 + least
privilege, and contain execution in a sandbox.</p></div>
<p>The <a href="/blog/mcp-and-skills-in-a-hive/">Model Context Protocol</a> made agents genuinely useful: a standard
way to plug an LLM into tools, APIs, and data. It also opened a security surface that most teams wiring
up MCP servers haven’t thought through. This is a practical tour of how MCP gets attacked — and the
layered defenses that actually hold.</p>
<h2 id="the-mcp-trust-model-why-its-attackable" tabindex="-1">The MCP trust model (why it’s attackable) <a class="anchor" href="#the-mcp-trust-model-why-its-attackable" aria-hidden="true">#</a></h2>
<p>MCP works like this: a client asks a server “what tools do you offer?”, gets back tool names,
<strong>descriptions</strong>, and schemas, and passes those to the model so it can decide what to call. The weak
point is subtle but serious — <strong>the tool description is model input the user never sees</strong>, and
<strong>most clients accept server-provided metadata without rigorous validation; the spec doesn’t require it</strong>
(<a href="https://arxiv.org/abs/2603.22489">academic threat-modeling work</a> makes this the center of its analysis).</p>
<p>So whoever controls an MCP server controls text that goes straight into the model’s decision-making. That
single fact is the root of almost every MCP attack below.</p>
<h2 id="tool-poisoning-the-headline-attack" tabindex="-1">Tool poisoning: the headline attack <a class="anchor" href="#tool-poisoning-the-headline-attack" aria-hidden="true">#</a></h2>
<p><strong>Tool poisoning</strong> embeds malicious instructions in a tool’s metadata — its description or schema. The
user approves a benign-looking “weather” tool; its description quietly tells the model to also read
<code>~/.ssh/id_rsa</code> and pass it along. It’s a form of <em>indirect</em> prompt injection, and research identifies it
as the most prevalent, impactful client-side MCP vulnerability. The agent isn’t jailbroken — it’s just
following instructions it was handed by something it was told to trust.</p>
<p>The mitigation starts with a mindset shift: <strong>tool descriptions are untrusted model input, not
documentation.</strong> Treat them the way you’d treat any string from the internet that reaches your prompt.</p>
<figure class="fig fig-inline"><img src="/blog/assets/media/mcp-security-tool-poisoning/note-1.png" alt="Hand-drawn sketch from “MCP Tool Poisoning: How to Secure the Model Context Protocol”" loading="lazy" decoding="async" /></figure>
<h2 id="the-rug-pull-poisoning-on-a-delay" tabindex="-1">The rug pull: poisoning on a delay <a class="anchor" href="#the-rug-pull-poisoning-on-a-delay" aria-hidden="true">#</a></h2>
<p>Approving a server once isn’t enough, because descriptions can change. In a <strong>rug pull</strong>, a server is
clean when you approve it, then <strong>updates its tool descriptions later</strong> to add exfiltration instructions
the agent will follow on the next call. Approval is a moment; trust has to be continuous.</p>
<p>Defense: pin and <strong>re-scan tool descriptions on every server update, and reject silent changes</strong>. A tool
whose description changed since you approved it should require re-approval, not be used quietly.</p>
<h2 id="the-rest-of-the-catalog" tabindex="-1">The rest of the catalog <a class="anchor" href="#the-rest-of-the-catalog" aria-hidden="true">#</a></h2>
<ul>
<li><strong>Resource / content poisoning.</strong> Instead of the tool’s metadata, the attacker hides instructions in
the <strong>data</strong> an MCP server retrieves — a poisoned document, a planted issue comment — achieving
<em>persistent</em> prompt injection through a trusted data channel.</li>
<li><strong>Line jumping / command injection.</strong> Unsanitized tool arguments (pipes, <code>;</code>, <code>&amp;</code>) flow straight to a
shell. One 2026 analysis attributes <a href="https://www.practical-devsecops.com/mcp-security-best-practices/">roughly 43% of MCP CVEs</a>
to command injection — the unglamorous bug that does real damage. Never interpolate model output
straight into a command.</li>
<li><strong>Confused deputy.</strong> A proxy MCP server acts with <em>its</em> privileges instead of the <em>user’s</em>. The
<a href="https://modelcontextprotocol.io/docs/tutorials/security/security_best_practices">spec’s mitigation</a> is
a per-user registry of approved <code>client_id</code>s, checked before any third-party authorization flow, with
consent stored securely.</li>
</ul>
<h2 id="the-boring-truth-its-usually-the-token" tabindex="-1">The boring truth: it’s usually the token <a class="anchor" href="#the-boring-truth-its-usually-the-token" aria-hidden="true">#</a></h2>
<p>Worth saying plainly, because it cuts against the spy-movie framing: industry analyses put <strong>token
mismanagement — a leaked or over-privileged API key — as the most common real-world MCP breach vector</strong>,
ahead of clever injection chains. The flashy attack gets the write-ups; the credential left in scope does
the damage. If you do one thing, make it <strong>least privilege</strong>: scope every tool’s token so it can only
touch the data it explicitly needs.</p>
<figure class="fig fig-inline"><img src="/blog/assets/media/mcp-security-tool-poisoning/note-2.png" alt="Hand-drawn sketch from “MCP Tool Poisoning: How to Secure the Model Context Protocol”" loading="lazy" decoding="async" /></figure>
<h2 id="defenses-layered" tabindex="-1">Defenses, layered <a class="anchor" href="#defenses-layered" aria-hidden="true">#</a></h2>
<p>No single control is enough; MCP security is defense in depth.</p>
<ul>
<li><strong>Treat tool metadata as untrusted.</strong> Scan descriptions at install <strong>and on every update</strong>; reject
silent changes (the rug-pull defense).</li>
<li><strong>Sanitize tool arguments</strong> before they reach a shell or API. Model output is never a safe command.</li>
<li><strong>AuthN/AuthZ + least privilege.</strong> Modern MCP standardizes on <strong>OAuth 2.1</strong> for HTTP transports
(replacing custom auth and raw API keys), with Dynamic Client Registration; scope tokens tightly. The
<a href="https://genai.owasp.org/resource/a-practical-guide-for-secure-mcp-server-development/">OWASP guide to secure MCP server development</a>
is the checklist to start from.</li>
<li><strong>Mitigate confused-deputy</strong> with per-user <code>client_id</code> registries and stored consent for proxies.</li>
<li><strong>Vet and pin servers.</strong> Prefer first-party or local servers; pin versions; review on update.</li>
<li><strong>Contain execution.</strong> Even a perfectly-authed call can be hijacked by poisoned content, so pair MCP
hygiene with an execution sandbox and <strong>default-deny network egress</strong> — the same defense-in-depth
posture covered in <a href="/blog/agent-security-and-sandboxing/">running AI agents safely</a>. The protocol layer
and the infrastructure layer reinforce each other.</li>
</ul>
<p><a href="https://www.redhat.com/en/blog/model-context-protocol-mcp-understanding-security-risks-and-controls">Red Hat’s overview</a>
and the <a href="https://modelcontextprotocol.io/docs/tutorials/security/security_best_practices">MCP spec’s own security best practices</a>
are the authoritative starting points to go deeper.</p>
<h2 id="securing-mcp-in-a-hive" tabindex="-1">Securing MCP in a hive <a class="anchor" href="#securing-mcp-in-a-hive" aria-hidden="true">#</a></h2>
<p>A <a href="/#what">multi-agent harness</a> gives every agent MCP tools and skills — which means it inherits this
whole threat surface, but also a local-first advantage in defending it. The trust problem shrinks when
the servers are <em>yours</em>: prefer <strong>local or first-party MCP servers</strong> (no third party controlling your
tool descriptions), keep <strong>prod credentials out of agents’ reach</strong> and tokens least-privileged, and
<strong>version-pin</strong> what you install. Layer that with what a hive already gives you — risky actions gated by
<a href="/blog/human-in-the-loop-approving-ai-agents/">human-in-the-loop approval</a> and every call recorded to an
<a href="/blog/append-only-event-log-agents/">append-only audit log</a> — and a poisoned tool has far fewer places
to do damage.</p>
<h2 id="the-takeaway" tabindex="-1">The takeaway <a class="anchor" href="#the-takeaway" aria-hidden="true">#</a></h2>
<p>MCP didn’t invent prompt injection; it gave it a standardized, high-trust delivery channel. The fix isn’t
to avoid MCP — it’s to stop trusting tool metadata by default. Treat descriptions as untrusted input,
scan and pin them, scope every token to least privilege, sanitize what reaches a shell, and contain
execution. Do that and MCP is what it should be: a powerful, <em>bounded</em> way to give agents real tools.</p>
<p>Want a hive where MCP tools run with least-privilege, human-gated, fully-audited by default? You can
<a href="/#install">download Munder Difflin</a> free — it’s open source.</p>


      
      <section class="faq" aria-label="Frequently asked questions">
        <h2 id="faq">FAQ</h2>
        
        <div class="qa">
          <p class="q">What is MCP tool poisoning?</p>
          <p class="a">Tool poisoning is an attack where malicious instructions are hidden in an MCP tool&#39;s metadata — its name, description, or schema. Because an MCP client passes those descriptions to the model as input the user never sees, a poisoned description can steer the agent into unsafe actions. It&#39;s a form of indirect prompt injection that exploits the client-server trust model.</p>
        </div>
        
        <div class="qa">
          <p class="q">What is an MCP rug pull attack?</p>
          <p class="a">A rug pull is a time-delayed tool poisoning. A server looks benign when you approve it, then later updates its tool descriptions to add malicious instructions — which the agent will follow on the next call. The defense is to treat tool descriptions as untrusted: scan them at install and on every update, and reject silent changes.</p>
        </div>
        
        <div class="qa">
          <p class="q">What&#39;s the most common MCP security risk?</p>
          <p class="a">Despite the attention tool poisoning gets, industry analyses point to something more mundane: token mismanagement — a leaked or over-privileged credential — as the most common real-world MCP breach vector. Least privilege and secret hygiene matter more than defending any single clever injection chain.</p>
        </div>
        
      </section>
      

      
      <hr />
      <div class="share">
        <span class="lbl">Tags</span>
        <a class="tag" href="/blog/tags/security/">Security</a><a class="tag" href="/blog/tags/mcp/">MCP</a><a class="tag" href="/blog/tags/protocols/">Protocols</a><a class="tag" href="/blog/tags/guardrails/">Guardrails</a>
      </div>
      
    </div>

    
    <aside class="toc-rail" aria-label="Table of contents">
      <div class="tt">On this page</div>
      <ol>
        <li class="lvl-2"><a href="#the-mcp-trust-model-why-its-attackable">The MCP trust model (why it’s attackable)</a></li><li class="lvl-2"><a href="#tool-poisoning-the-headline-attack">Tool poisoning: the headline attack</a></li><li class="lvl-2"><a href="#the-rug-pull-poisoning-on-a-delay">The rug pull: poisoning on a delay</a></li><li class="lvl-2"><a href="#the-rest-of-the-catalog">The rest of the catalog</a></li><li class="lvl-2"><a href="#the-boring-truth-its-usually-the-token">The boring truth: it’s usually the token</a></li><li class="lvl-2"><a href="#defenses-layered">Defenses, layered</a></li><li class="lvl-2"><a href="#securing-mcp-in-a-hive">Securing MCP in a hive</a></li><li class="lvl-2"><a href="#the-takeaway">The takeaway</a></li>
      </ol>
    </aside>
    
  </div>

  <footer class="post-foot wrap"><div class="prevnext">
      <a class="pn prev" href="/blog/multi-agent-orchestration-patterns/"><span class="k">← Older</span><span class="t">Multi-Agent Orchestration Patterns: Which Topology, and When</span></a>
      <a class="pn next" href="/blog/how-to-install-and-use-munder-difflin/"><span class="k">Newer →</span><span class="t">How to Install Munder Difflin 0.5.6: A Beginner&#39;s Guide From Zero to a Working Office</span></a>
    </div>
    <section class="related">
      <h2>Related in Guides</h2>
      <div class="post-grid">
        <article class="card t-guides">
  <a class="thumb t-guides" href="/blog/claude-md-best-practices/" aria-hidden="true" tabindex="-1"><img src="/blog/assets/media/claude-md-best-practices/hero.png" alt="" loading="lazy" decoding="async" /></a>
  <div class="body">
    <div class="meta">
      <time datetime="2026-10-26">Oct 26, 2026</time>
      <span class="dot" aria-hidden="true"></span>
      <span>4 min</span>
    </div>
    <h3><a href="/blog/claude-md-best-practices/">CLAUDE.md Best Practices: What to Write, Where It Lives, How Long (Oct 2026)</a></h3>
    <p class="dek">CLAUDE.md best practices from Anthropic&#39;s docs, checked 26 Oct 2026: what to put in it, where each file lives, the 200 line target and how AGENTS.md loads.</p>
    <div class="foot">
      <span class="tag kind">Guides</span>
      <a class="more" href="/blog/claude-md-best-practices/" aria-label="Read: CLAUDE.md Best Practices: What to Write, Where It Lives, How Long (Oct 2026)">Read →</a>
    </div>
  </div>
</article>
<article class="card t-guides">
  <a class="thumb t-guides" href="/blog/codex-computer-use/" aria-hidden="true" tabindex="-1"><img src="/blog/assets/media/codex-computer-use/hero.png" alt="" loading="lazy" decoding="async" /></a>
  <div class="body">
    <div class="meta">
      <time datetime="2026-10-26">Oct 26, 2026</time>
      <span class="dot" aria-hidden="true"></span>
      <span>4 min</span>
    </div>
    <h3><a href="/blog/codex-computer-use/">Codex Computer Use: How to Turn It On, Plans and Limits (Oct 2026)</a></h3>
    <p class="dek">Codex computer use lets the ChatGPT desktop app see and click apps on macOS and Windows. Setup, which plans get it, locked use and how Claude&#39;s compares.</p>
    <div class="foot">
      <span class="tag kind">Guides</span>
      <a class="more" href="/blog/codex-computer-use/" aria-label="Read: Codex Computer Use: How to Turn It On, Plans and Limits (Oct 2026)">Read →</a>
    </div>
  </div>
</article>
<article class="card t-guides">
  <a class="thumb t-guides" href="/blog/cursor-pricing/" aria-hidden="true" tabindex="-1"><img src="/blog/assets/media/cursor-pricing/hero.png" alt="" loading="lazy" decoding="async" /></a>
  <div class="body">
    <div class="meta">
      <time datetime="2026-10-26">Oct 26, 2026</time>
      <span class="dot" aria-hidden="true"></span>
      <span>5 min</span>
    </div>
    <h3><a href="/blog/cursor-pricing/">Cursor Pricing (Oct 2026): Every Plan After the 12 Oct Change</a></h3>
    <p class="dek">Cursor pricing as of 26 Oct 2026: Free, Plus $8, Premium $30, Super $100, Ultra $200, Teams seats, what changed on 12 Oct and what old plans keep.</p>
    <div class="foot">
      <span class="tag kind">Guides</span>
      <a class="more" href="/blog/cursor-pricing/" aria-label="Read: Cursor Pricing (Oct 2026): Every Plan After the 12 Oct Change">Read →</a>
    </div>
  </div>
</article>

      </div>
    </section>
    
  </footer>
</article>

</main>
<footer class="site-foot">
  <div class="wrap">
    <div class="foot-top">
      <a class="foot-brand brand" href="https://munderdiffl.in" aria-label="Munder Difflin home">
        <img class="brand-logo" src="https://munderdiffl.in/logo.png" width="1318" height="840" alt="Munder Difflin, Inc. — Multi-Agent Harness" />
      </a>
      <div class="foot-links">
        <a href="/blog/">Latest</a>
        <a href="/blog/topics/">Topics</a>
        <a href="/blog/about/">About</a>
        <a href="https://munderdiffl.in/blog/feed.xml">RSS</a>
        <a href="https://github.com/chaitanyagiri/munder-difflin">GitHub</a>
        <a href="https://munderdiffl.in">munderdiffl.in ↗</a>
      </div>
    </div>
    <div class="foot-meta">
      © 2026 Munder Difflin · Local-first · MIT-licensed ·
      Built on <a href="https://munderdiffl.in">Electron, React, Pixi.js, xterm.js &amp; node-pty</a>.
      Run an office of agents while you sleep.
    </div>
  </div>
</footer>

<script>
  // sticky-nav border on scroll
  (function () {
    var nav = document.getElementById('nav');
    if (!nav) return;
    var onScroll = function () { nav.classList.toggle('scrolled', window.scrollY > 8); };
    onScroll(); window.addEventListener('scroll', onScroll, { passive: true });
  })();
  // YouTube click-to-load: thumbnail button → youtube-nocookie iframe
  (function () {
    document.addEventListener('click', function (e) {
      var btn = e.target.closest('.yt-load');
      if (!btn) return;
      var box = btn.closest('.yt');
      var id = box && box.getAttribute('data-yt');
      // Only an 11 character YouTube video id may reach the iframe URL.
      if (!id || !/^[A-Za-z0-9_-]{11}$/.test(id)) return;
      var frame = document.createElement('iframe');
      frame.src = 'https://www.youtube-nocookie.com/embed/' + encodeURIComponent(id) + '?autoplay=1';
      frame.title = box.getAttribute('data-title') || 'YouTube video';
      frame.allow = 'accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share';
      frame.setAttribute('allowfullscreen', '');
      box.replaceChildren(frame);
    });
  })();
</script>

</body>
</html>