<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[PhysioAI Engineering]]></title><description><![CDATA[PhysioAI Engineering]]></description><link>https://physioai.hashnode.dev</link><generator>RSS for Node</generator><lastBuildDate>Mon, 16 Nov 2026 18:45:38 GMT</lastBuildDate><atom:link href="https://physioai.hashnode.dev/rss.xml" rel="self" type="application/rss+xml"/><language><![CDATA[en]]></language><ttl>60</ttl><item><title><![CDATA[How we put a physiotherapy AI tutor inside Claude and ChatGPT with MCP]]></title><description><![CDATA[A small, read-only MCP server that answers study questions and ends every answer with a practice MCQ]]></description><link>https://physioai.hashnode.dev/how-we-put-a-physiotherapy-ai-tutor-inside-claude-and-chatgpt-with-mcp</link><guid isPermaLink="true">https://physioai.hashnode.dev/how-we-put-a-physiotherapy-ai-tutor-inside-claude-and-chatgpt-with-mcp</guid><category><![CDATA[mcp]]></category><category><![CDATA[AI]]></category><category><![CDATA[healthtech]]></category><category><![CDATA[chatgpt]]></category><category><![CDATA[claude]]></category><dc:creator><![CDATA[Maitreya Hiware]]></dc:creator><pubDate>Sun, 15 Nov 2026 19:10:14 GMT</pubDate><content:encoded><![CDATA[<p>Physiotherapy students already ask general chatbots their study questions. The answers are usually fine, but they are generic: no exam framing, no red flags, and nothing that checks whether the student actually understood. At <a href="https://physioai.in">PhysioAI</a> we wanted our tutor, Synergist, to show up where students already are, so we built a Model Context Protocol (MCP) connector for it.</p>
<p>This post covers what the connector does, how it is shaped, and a few decisions that kept it simple.</p>
<h2>What it does</h2>
<p>Ask a physiotherapy question in Claude or ChatGPT, and PhysioAI answers it the way an exam tutor would:</p>
<ul>
<li>A structured answer with key points, assessment details and red flags</li>
<li>One exam-style multiple-choice question at the end, with the correct answer and a short explanation</li>
<li>Two modes: <strong>Rapid</strong> for quick revision (the default) and <strong>Deep Study</strong> for a full, phase-by-phase explanation</li>
</ul>
<p>It covers anatomy, biomechanics, musculoskeletal, neurological, cardiorespiratory, paediatric and sports rehab, plus licensing exams such as the NPTE, DHA and Prometric.</p>
<h2>Two tools, both read-only</h2>
<p>We kept the surface area deliberately small:</p>
<table>
<thead>
<tr>
<th>Tool</th>
<th>What it does</th>
</tr>
</thead>
<tbody><tr>
<td><code>ask_synergist</code></td>
<td>Answers a study question (rapid or deep) and returns the answer plus one practice MCQ</td>
</tr>
<tr>
<td><code>get_profile</code></td>
<td>Returns the connected account's display name and email</td>
</tr>
</tbody></table>
<p>Neither tool creates, changes or deletes anything. A study tutor does not need write access, and keeping both tools read-only makes the connector easy to reason about and easy to approve in a client.</p>
<h2>Transport and auth</h2>
<p>The server speaks Streamable HTTP and uses OAuth 2.1 with PKCE:</p>
<pre><code>https://auth.physioai.in/functions/v1/synergist-mcp/mcp
</code></pre>
<p>When you click <strong>Connect</strong>, the client opens the PhysioAI sign-in page. You sign in with email, Google or Apple, and the client gets a token. Your password never reaches Claude or ChatGPT, and the server only uses your account ID, name and email to identify you.</p>
<p>A fair-use limit of 50 questions per account per day keeps costs predictable. It resets daily at 00:00 UTC.</p>
<h2>Connecting it</h2>
<p><strong>Claude:</strong> Settings → Connectors → add PhysioAI (or <em>Add custom connector</em> with the URL above) → Connect → sign in.</p>
<p><strong>ChatGPT:</strong> Apps → PhysioAI → Connect → sign in, then mention <strong>@PhysioAI</strong> in a chat. Answers render in a card with an interactive MCQ.</p>
<p>Prompts that work well:</p>
<ul>
<li>"Explain the Lachman test and how it differs from the anterior drawer test"</li>
<li>"Deep Study: ACL reconstruction rehab, phase by phase"</li>
<li>"Quick NPTE revision: Brunnstrom stages of stroke recovery"</li>
</ul>
<h2>What we learned</h2>
<ol>
<li><strong>End every answer with a question.</strong> The practice MCQ turned a passive answer into active recall, and it is the part students mention most.</li>
<li><strong>Default to short.</strong> Most questions are quick revision. Making Rapid the default and Deep Study opt-in kept answers readable in a chat window.</li>
<li><strong>Stay educational.</strong> The connector explains concepts. It does not diagnose or advise on real patients, and it says so.</li>
</ol>
<h2>Try it</h2>
<p>The full connector docs are at <a href="https://physioai.in/ai-apps">physioai.in/ai-apps</a>. PhysioAI itself, with Patient Simulator, licensing exam mock tests and Clinic OS, is at <a href="https://physioai.in">physioai.in</a>.</p>
]]></content:encoded></item></channel></rss>