05 / AI operations / APIs
Safer infrastructure checks with named tools
A typed Model Context Protocol service that turns common API and SSH checks into smaller operations with clear names and inputs.
Starting point
Where it started
Broad shell access makes routine infrastructure work difficult to constrain, explain, and audit.
What I did
What I worked on
- Created named tools for virtualization, firewall, proxy, mail, monitoring, and service checks.
- Defined typed inputs and predictable responses for each operation.
- Separated credentials from code and kept public descriptions sanitized.
- Paired the tools with source-of-truth documentation and outcome verification.
Result
What changed
Common infrastructure checks now start from explicit intent, creating a clearer boundary between a request and the system underneath it.
What I learned
What I took from it
- A named tool is easier to review than an arbitrary command.
- Read-only and mutating actions should be visibly distinct.
- Tools still need current context and post-action verification.
Next case study / 06