https://meshrefine.com/en/tags/oauth/Posts · Mesh RefinementThu, 19 Jun 2025 20:14:35 +0200https://meshrefine.com/en/posts/mcp_update_18_06_2025/<p>The Model Context Protocol, despite its aggressive adoption (or perhaps because of it), continues to evolve. <a href="https://modelcontextprotocol.io/specification/2025-06-18/changelog">Anthropic recently updated the MCP specification</a>, and below, we&rsquo;ll look at the main changes.</p> <h2 id="0c0bef0c025560d849b4de6b55b2f23e-security-enhancements">Security Enhancements</h2> <p>An MCP server is now always classified as an <code>OAuth Resource Server</code>, and clients are required to implement <a href="https://www.rfc-editor.org/rfc/rfc8707.html">Resource Indicators (RFC 8707)</a>. This is necessary to protect against attacks like the <a href="https://en.wikipedia.org/wiki/Confused_deputy_problem">Confused Deputy</a>. Previously, tokens requested by a client from an authorization server were &ldquo;impersonal,&rdquo; meaning they could be used by anyone. This allowed an attacker to create a phishing MCP server, deceive a client, steal the token, and use that token to gain access to the real MCP server.</p>Thu, 19 Jun 2025 20:14:35 +0200https://meshrefine.com/en/posts/mcp_update_18_06_2025/