Use the Olostep CLI with AI agents

AI coding agents can run olostep like any other command. The CLI notices when a coding agent runs it and behaves for a program instead of a person.

Set it up

Install the CLI and sign in yourself

curl -fsSL https://cli.olostep.com | sh
olostep login

The agent uses your sign-in on this computer. In CI, set OLOSTEP_TOKEN instead. See Authentication.

Give the agent the CLI guide

olostep skill get olostep-cli --write .claude/skills

This writes .claude/skills/olostep-cli/SKILL.md: how to run the CLI, its exit codes, and every command with its flags and JSON output schema. Use your agent’s skills folder in place of .claude/skills.

Tell the agent to use --json

Agents should run commands with --json and read the JSON. The human output is for people and can change.

How the CLI behaves for an agent

The CLI treats a run as an agent’s when there is no terminal, or when one of CLAUDECODE, CURSOR_AGENT, CODEX_CI or AI_AGENT is set. Then:

olostep with no command prints help, not the welcome screen.
It never prompts. A command that needs sign-in exits with code 3 and tells the agent, on stderr, to run olostep login or set OLOSTEP_TOKEN.
A command that may delete or overwrite something exits with code 4 unless you pass --yes.
olostep keys reveal never prints a key. Use --write-env <file> to save it to a git-ignored env file.
olostep chat prints replies as plain text (--plain is on), and needs a message.
No update notice is printed.

Exit code	What the agent should do
`0`	Read the output.
`1`	Read the error on stderr.
`2`	Check the command with `olostep <command> --help`.
`3`	Ask the person to run `olostep login`.
`4`	Confirm with the person, then run again with `--yes`.
`5`	The server could not be reached. Retry later.

Find the right command

olostep help monitor pause        # commands whose name or description has every word
olostep tools --search crawl      # tools that mention "crawl"
olostep tools crawl_create        # one tool's full input and output schema

`olostep skill`

Skills are short instructions for one kind of task. List them, read them, or write them into an agent’s skills folder:

olostep skill list
olostep skill get credit-usage
olostep skill get olostep-cli dashboard --write .claude/skills
olostep skill get olostep-cli --json

Skill	What it covers
`olostep-cli`	How to use the `olostep` command line. Generated from the CLI’s commands.
`credit-usage`	Credit balances, plans, credit expiry and daily usage.
`dashboard`	Dashboard features: account, billing, team, API keys, logs, jobs, parsers, monitors, agents, support and integrations.

olostep skill list shows the current list. See olostep skill for the flags.

`olostep tools`

Every CLI command calls a tool on the Olostep server. olostep tools lists them, and marks the ones that change things:

olostep tools
olostep tools --json

Give feedback

When a command fails, is missing, or is confusing, the agent can tell the Olostep team. This works without signing in:

olostep feedback --about tooling --message "crawl create rejected --depth 0"

Use --about tooling for the CLI itself and --about product for Olostep. Never include keys, passwords or personal details.

MCP instead of a terminal

If your agent supports MCP, you can connect it to Olostep directly. See MCP server, or print the setup for your client:

olostep integrations config --client claude_code

​Set it up

​How the CLI behaves for an agent

​Find the right command

​olostep skill

​olostep tools

​Give feedback

​MCP instead of a terminal