# Getting started

Pick the AI agent or environment where you want to use Slideshot. Each setup page covers the three integration paths (MCP server, skill, and CLI) and includes a copyable instruction you can paste into the agent so it can walk you through the rest.

## Choose your agent

- [Claude Code & Cowork](/docs/getting-started/claude-code) - Set up Slideshot for Claude Code (CLI path) or Cowork (plugin marketplace path).
- [Codex](/docs/getting-started/codex) - Connect the Slideshot MCP server or CLI to the OpenAI Codex terminal agent.
- [ChatGPT](/docs/getting-started/chatgpt) - Use the Slideshot skill or call the API directly from custom GPTs and Projects.
- [Cursor](/docs/getting-started/cursor) - Add the Slideshot MCP server to Cursor and let the editor agent drive recordings.
- [Other agents](/docs/getting-started/other-agents) - Generic setup flow for any agent that supports MCP, shell access, or HTTP requests.

## What you will set up

Each setup page covers up to three pieces. You usually only need one: pick the path that fits the agent you chose above.

- **MCP server**: Connect the agent to `https://api.slideshot.ai/mcp`. The agent can then call Slideshot tools like `create_run`, `get_run`, `list_run_artifacts`, and `submit_run_input` directly from the conversation.
- **Skill**: A short instruction file that teaches the agent how to plan a Slideshot recording: when to ask for credentials, how to choose run options, and how to handle paused runs that ask for OTP input.
- **CLI**: The `slideshot` npm package. Useful for terminal-driven workflows, scripts, and CI jobs even when the agent also has MCP access.

## Before you start

- A Slideshot account and an [API key](/docs/api).
- The URL of the web app you want to record.
- Any login credentials the target app needs. Save them up front with [credentials](/docs/api/credentials) so the agent does not have to handle them mid-run.

## Not sure which path to choose

- Use **MCP** when you want the agent to drive recording from natural-language conversation. It is the lowest-friction path inside Claude Code, Codex, Cursor, and any other MCP-capable agent.
- Use the **CLI** when you want a scriptable command for CI jobs, release automation, or scheduled product marketing workflows.
- Use the **skill** alongside either path so the agent makes good defaults. For example, knowing when to set `artifacts.gif=true` for a social clip or when to enable `video.blur_emails` for a public demo.

See [MCP versus CLI](/docs/mcp-vs-cli) for a longer comparison.