SysKnife — Claude Code MCP server: what it does, how to install
Runs Linux administration tasks through typed actions with terminal approval receipts. Helps when an assistant needs system access without exposing a raw shell.
Open on GitHubWhat SysKnife does
SysKnife replaces direct shell access with strict, typed system procedures. The assistant proposes changes by calling sysknife_plan to inspect risk scores and generate transaction identifiers. The daemon constructs the underlying system calls itself. This architecture prevents prompt injections from hiding malicious flags inside shell text.
Privileged actions cannot run automatically. The model pauses execution once it produces a transaction identifier. To proceed, the user enters sysknife approve <transaction-id> directly in a separate terminal window. This creates a one-time cryptographic receipt for sysknife_execute. An Ed25519-signed journal logs every operation to prevent tampering.
The setup wizard requires Node 22 or newer. Default packages on distributions like Ubuntu 22.04 supply older runtimes and cause the setup process to stop immediately. On hosts using rpm-ostree, failed changes revert to their earlier state automatically.
What it helps with
- replace raw shell string execution with structured typed actions
- enforce out-of-band user approval for dangerous system operations
- track administrative tasks in an Ed25519-signed verification log
- roll back failed package updates automatically on atomic hosts
Example prompts for the agent
- Plan opening port 80 with UfwAllow and show the risk level
- Check disk space usage using sysknife tools
- Inspect recent sysadmin runs and verify the cryptographic audit chain
How to install SysKnife
Install
The MCP registry has no launch command for it: see the repository README
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about SysKnife
How to install SysKnife in Claude Code?
Run npx sysknife-setup in your terminal. The interactive script validates release hashes and registers the server in your configuration.
Why use SysKnife instead of giving an AI shell access?
AI models can bypass raw shell filters through prompt manipulation. SysKnife removes shell string input entirely and lets the daemon handle verified action structures.
How does manual approval work?
When an action requires permission, the agent stops and displays a transaction ID. You run sysknife approve <transaction-id> in your own terminal to yield an execution receipt.
