<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Gig&apos;MCP blog</title><description>Engineering notes on MCP security: sandboxing, credential injection, egress control, and building a security-first MCP gateway.</description><link>https://gigmcp.dev/</link><item><title>Gig&apos;MCP vs Composio: self-hosted security vs hosted convenience</title><link>https://gigmcp.dev/blog/gigmcp-vs-composio/</link><guid isPermaLink="true">https://gigmcp.dev/blog/gigmcp-vs-composio/</guid><description>Two very different answers to the same problem: how do you give AI agents access to hundreds of tools without scattering API keys everywhere? A fair comparison of Composio&apos;s hosted connector platform and Gig&apos;MCP&apos;s sandboxed, self-hosted gateway.</description><pubDate>Sun, 07 Jun 2026 00:00:00 GMT</pubDate></item><item><title>How Gig&apos;MCP keeps API keys out of MCP servers entirely</title><link>https://gigmcp.dev/blog/keeping-api-keys-out-of-mcp-servers/</link><guid isPermaLink="true">https://gigmcp.dev/blog/keeping-api-keys-out-of-mcp-servers/</guid><description>Placeholder tokens, an envelope-encrypted vault, and a MITM egress proxy that injects real credentials only on HTTPS calls to allowlisted domains. A technical walkthrough of credential injection at the network boundary.</description><pubDate>Sun, 07 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Why MCP servers are a security risk (and what kernel sandboxing fixes)</title><link>https://gigmcp.dev/blog/mcp-servers-security-risk/</link><guid isPermaLink="true">https://gigmcp.dev/blog/mcp-servers-security-risk/</guid><description>Every MCP server you install runs with your raw API keys and unrestricted network access. Here&apos;s the threat model the ecosystem is ignoring, and how kernel-enforced sandboxing with bubblewrap, namespaces, and seccomp actually closes it.</description><pubDate>Sun, 07 Jun 2026 00:00:00 GMT</pubDate></item></channel></rss>