This quick tutorial gets you from zero to your first completed task with Claude Code: set up, run a task, use plan mode, and extend it with tools. If you'd rather skip setup, you can spawn a hosted agent instead.
Key takeaways
- Start in a real project so the agent has code to work with.
- Give concrete goals, then use plan mode for larger multi-file changes.
- Connect MCP servers to let the agent act beyond the codebase.
1. Set up
Install the CLI and authenticate (see our install guide). Open Claude Code inside a real project so it has code to work with.
2. Run a task
Give it a concrete goal, "fix the failing test in checkout.ts". It explores, plans, edits, and runs the test, iterating until it passes.
3. Use plan mode
For bigger changes, let it produce a plan first and approve it before it edits, safer for multi-file work.
4. Add tools with MCP
Connect MCP servers (GitHub, Slack, Supabase…) so the agent can act beyond the codebase. Creative Tim's hosted agents come with these connectors preconfigured.
Skip the setup, spawn a hosted agent
Creative Tim hosts Claude-Code-compatible harnesses (OpenClaw, Hermes), preconfigured with connectors. Bring your own key or run on credits.
Frequently asked questions
Do I need to be an expert?
No, you describe goals in plain English. Start small and review its changes.
What's the fastest way to try it?
Spawn a hosted, preconfigured agent rather than installing the CLI locally.
What should my first task be?
Pick something small and verifiable, like fixing a failing test or adding a helper, so you can review the changes with confidence.
Keep reading