Make MCP executable recover on non-FHS Linux

[st0012]

Context

rubydex_mcp now ships as a precompiled Rust executable inside the gem and is launched through the Ruby wrapper at exe/rubydex_mcp. On normal platforms, the wrapper can exec that packaged binary directly.

Nix-style Linux environments can break that direct launch. The binary is present and executable, but the kernel may still raise ENOENT because the ELF interpreter path baked into the binary, for example /lib64/ld-linux-x86-64.so.2, does not exist in the host filesystem. From an MCP client, this looks like the server cannot start even though the gem installed correctly.

Change

This keeps the normal path unchanged: the wrapper still tries to exec the packaged MCP binary first.

If that direct exec raises Errno::ENOENT, the wrapper now looks for the dynamic loader already mapped into the current Ruby process through /proc/self/maps. When it finds that loader, it re-execs the packaged MCP binary through it.

The launcher logic now lives in Rubydex::MCPServerLauncher, keeping exe/rubydex_mcp small and keeping these process-launch details out of the main Ruby API.

[st0012]

• Make MCP executable recover on non-FHS Linux #840 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.