Skip to content

Quickstart

First run of the ai_stp CLI and the shortest path to a checked environment.

Quickstart

Install the CLI

The CLI installs as an ordinary uv tool:

uv tool install ai-stp-cli

Then check that the command is available:

ai-stp version
ai-stp doctor --json

doctor reports the state of the environment, what this installation can do, and anything that has to be fixed before assembling a setup.

Prepare the device

For local work without an account, create a developer passport and a device:

ai-stp passport developer init --json
ai-stp device init --json

The device gets a device_id and a key. The private key is held in the operating system's secret store; where that is unavailable, the CLI says plainly that it fell back to file storage.

Look at the catalog

Anonymous reading of the public catalog needs no sign-in:

ai-stp registry search --json

For one exact object, use show. With no network, the CLI may answer from the last confirmed local copy and will say when the platform last confirmed it.

What comes next

After the first doctor, the agent reads the machine help:

ai-stp help --agent --json

That help is the source of commands for the Agent Skill. An agent should not invent commands from memory when the CLI already answers with them.

=== "I am choosing a published setup" Open the catalog and check the trust line, the compatibility with your harness, and what the setup is made of.

=== "I am assembling my own setup" Start with components: it helps decide what belongs in a skill, what in an mcp, and what stays a setting or an instruction.

=== "I am publishing a component" See publishing and the trust rules. A published version is immutable, so it is worth checking the passport before the first release.

ai_stp