rtk-mcp: a Rust MCP server that runs commands through RTK

[admiralhr99]

I maintain rtk-mcp, an MIT-licensed MCP server in Rust. Its run_command tool runs a shell command through rtk rewrite, so an MCP client's model reads RTK-filtered output. It also has tools for RTK's status, savings (rtk gain), discover, session, recall, config and rewrite, plus install_rtk, which installs or upgrades RTK after verifying the release's SHA-256 checksum.

• Repo: https://github.com/admiralhr99/rtk-mcp
• It started from Ousama Ben Younes's original rtk-mcp; most of the new code was written with Claude Code.
• Tested with RTK 1.1.0 and rtk-mcp 0.8.1 through the Claude desktop app's remote devices on Linux. CI runs on Linux, macOS and Windows. Not affiliated with rtk-ai.

Two things from testing that might be useful here:

1. Through that bridge, MCP tools arrive deferred (name only), and the anthropic/alwaysLoad hint did not take effect in my test. One line in my personal preferences worked: use run_command, and if only its name is listed, load it first with tool search. The README documents this under "Getting the model to use it". It may matter for anyone exposing RTK to Claude Desktop as MCP tools instead of hooks.
2. As I understand feat(desktop-plugin): Claude Desktop plugin, release job and docs #4967, the new Claude Desktop plugin avoids that by compressing the output of tools the model already uses. rtk-mcp is for clients without a hook point, so I see it as complementary.

Question: would you be open to listing community integrations like this, for example in a short "Community" section on the Supported Agents page? If so, I'll send a small docs PR (after signing the CLA). If you'd rather not link external projects, that's fine. Thanks for RTK.