Generate an Agent-Native CLI From Any API or Website

In this guide, you will learn how to use Printing Press (https://printingpress.dev/) to download agent-friendly tools for hard to use APIs like Google Flights or ESPN. You will also learn to build your own CLI tools to access any site with an API. We use Codex but this works great with OpenClaw and Claude Code too.

  • Beginner
  • 5 steps
  • By The Rundown University

Rundown Pro

Every course and recording, the certification program and member perks, for $29 a month. Guides stay free.

Go Pro for $29

Step by step 5 steps

  1. Step 1 of 5

    Start With Printing Press

    Open the Printing Press website (https://printingpress.dev/) and skim the examples before you install anything. The point is not just that Printing Press creates CLIs. The useful part is that it creates agent-friendly tools: compact commands, help output, and repeatable workflows your agent can call without browsing around every time.

    You can also read the Printing Press GitHub repo (https://github.com/mvanhorn/cli-printing-press) for install notes and implementation details. Keep that link handy. In our test, giving the agent the relevant GitHub page helped it recover when it did not know how a starter skill was supposed to be configured.

  2. Step 2 of 5

    Install the Starter Kit

    Install the starter kit first. This gives your agent working examples before you ask it to generate anything new.

    Run:

    npx -y @mvanhorn/printing-press install starter-pack
    Printing Press starter kit install command
    Printing Press starter kit install command

    The starter kit includes example tools such as ESPN and flight-goat. Those are useful because they let you prove the setup works before you start designing your own commands.

    Pro tip: do not debug installation and tool generation at the same time. First prove the starter kit works. Then move on to generating your own tools.

  3. Step 3 of 5

    Have Your Agent Test a Starter Tool

    Now ask your agent to inspect the starter kit and try one safe read-only command. Use this prompt:

    Inspect the starter kit. List the tools, read their help output, and try one safe read-only command from ESPN or flight-goat. Tell me which tool worked and which command you ran.

    For ESPN, the agent should be able to inspect the CLI help and run a command that pulls live or recent sports data. In our Codex run, the ESPN CLI returned completed NBA playoff games from ESPN data.

    ESPN CLI scoreboard commands
    ESPN CLI scoreboard commands

    If the agent gets stuck, point it back to the Printing Press website or GitHub repo and ask it to inspect the specific skill. Many generated skills include a doctor or verification command. That is the first thing to run when a starter tool does not work.

    Pro tip: make sure your agent has access to GitHub. Printing Press tools often need the repo context, install instructions, or skill files to finish setup cleanly.

  4. Step 4 of 5

    Install the Printing Press Binary

    Once the starter kit works, install the Printing Press binary so your agent can generate its own tools:

    go install github.com/mvanhorn/cli-printing-press/v4/cmd/printing-press@latest

    Then ask your agent to confirm the binary is available:

    Confirm printing-press is installed.

    Do not skip this check. If the binary is not available, your agent should fix that before it tries to generate a new CLI.

  5. Step 5 of 5

    Generate Commands for Your Own Website

    Start with a simple site that has a lot of public data. A docs site, public catalog, sports site, recipe site, or API-backed directory is easier than a complex app with private endpoints.

    Use this prompt:

    Use Printing Press to generate read-only commands for [WEBSITE].
    Propose the useful commands first, then generate and test them.

    Ask for read-only commands first. Good starter commands are usually search, lookup, compare, export, or summarize. Save write actions for later, after you know the generated tool is safe and accurate.

    If the site has an API key, docs page, or OpenAPI spec, give that to your agent. If it does not, ask the agent to look for public API endpoints and avoid private endpoints unless you explicitly provide credentials.

That's the whole workflow

Mark it complete to keep your progress. The background follows.

Keep learning with Rundown Pro

Guides stay free. This is what Rundown Pro adds:

  • Every course and recording
  • The Rundown Certified Operator program
  • Member perks

$29/mo, cancel anytime

Background and context

Who This Is For

  • Developers and AI operators who want agents to use websites and APIs without long browser sessions.
  • Codex, OpenClaw, Hermes, or Claude Code users who want better local tools for repeated workflows.
  • Anyone who wants to turn a data-rich website or API into reusable read-only commands.

What You Will Build

You will install the Printing Press starter kit, have your agent test a working tool like ESPN, then install the Printing Press binary so your agent can generate its own CLI commands.

ESPN CLI returning live NBA scores
ESPN CLI returning live NBA scores

What You Need

  • Node and npx for the starter kit install.
  • Go for the Printing Press binary install.
  • An AI coding agent such as Codex, OpenClaw, Hermes, or Claude Code.
  • A simple website, API docs page, or OpenAPI spec to test after setup.

Going Further

If your agent struggles to use a specific MCP or integration, you can tell it to build its own commands with Printing Press. You just need a site with useful data, API documentation, or an API key.

The bigger idea is to stop making agents rediscover the same surface area every time. If your agent needs the same website or API every week, turn that system into a small local tool.