Tools to Get Started Working With Agents

If you are just getting started, do not try to learn every agent tool at once.

Start with five components:

a harness to run the agent
a terminal if you are using CLI agents
a few tools the agent can use
a basic understanding of context management
- A simple way to manage reusable context is with project instruction files such as AGENTS.md.
one real workflow to test

This page lists the practical tools that matter first.

Harnesses

A harness is the environment where you run an agent. It gives the model instructions, access to files, and access to tools.

If you are using CLI agents, use a dedicated terminal.

Start with a subscription for simplicity. Use API billing later if you need shared team usage or automation.

These are the first tool categories to understand because they let agents gather information and take useful action.

Use this when the workflow happens inside a website.

Use this when the agent needs to find information first.

Use this when you want the underlying content from a page or API.

Use this when the workflow depends on leads, accounts, contacts, notes, or deal status.

Use this when you want the agent to work with your actual systems through custom tools.

Before going deeper, review Concepts, then Working with agents.

The most important ones to understand early are:

Harness: pick one GUI option (pi.dev, Claude Cowork, or Codex) or one CLI option (Claude Code or Codex CLI)
Terminal: Ghostty if you are using a CLI harness
Research: web search + web fetch
Action: web browser
Business system: CRM
Extension layer: custom tools or extensions
Concepts: context management, MCP, and skills