# Other agents setup

Any agent that supports MCP, shell access, or HTTP requests can drive Slideshot. The generic setup is to pick the path your agent supports: MCP first, CLI second, raw HTTP third: and load the Slideshot skill so the agent makes good defaults.

## Copy this into your agent

Paste this instruction into the agent and let it adapt to its environment.

```text
You are helping me connect Slideshot to this agent.

1. Detect what is available in this environment:
   - Can you connect to MCP servers? If yes, add the Slideshot server at
     https://api.slideshot.ai/mcp and walk me through the OAuth approval.
   - Can you run shell commands? If yes, you can call the Slideshot CLI
     through `npx slideshot <command>` with `SLIDESHOT_API_KEY` set in my shell.
   - Can you make outbound HTTP requests? If yes, you can call the Slideshot
     API directly with my API key.

2. Load the Slideshot skill from https://github.com/slideshot/skills so you
   know how to plan a recording, choose run options, and handle paused runs.

3. Create one small test recording against a public URL and confirm you can
   read the `demo.mp4` artifact when the run succeeds.

Do not print my API key or credentials back in any output.
```

## Pick a path

### MCP server

If your agent supports MCP, add a server with:

- **Name:** `slideshot`
- **Transport:** HTTP (streamable)
- **URL:** `https://api.slideshot.ai/mcp`

The first request triggers an OAuth approval flow in the browser. After approval, Slideshot tools appear in the agent's tool list:

- `create_run`, `get_run`, `list_runs`, `cancel_run`, `submit_run_input`, `list_run_artifacts`
- `list_credentials`, `create_credential`, `update_credential`, `set_default_credential`, `delete_credential`
- `submit_feedback`, `fetch_brand`

### CLI

If your agent can run shell commands, export your API key once and call the CLI through `npx`:

```bash
export SLIDESHOT_API_KEY="sk_live_xxxx"

npx slideshot runs create https://app.example.com \
  --goal "Open the analytics dashboard and create a new project"
```

### Raw HTTP

If your agent only makes outbound HTTP requests, point it at the [HTTP API](/docs/api) directly. The minimal call is:

```bash
curl -X POST https://api.slideshot.ai/v1/agent/runs \
  -H "x-api-key: $SLIDESHOT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "target_url": "https://app.example.com", "goal": "..." }'
```

## Load the skill

The Slideshot skill at [`github.com/slideshot/skills`](https://github.com/slideshot/skills) is plain markdown and works in any agent that loads instruction files or system prompts. It encodes recording defaults, awaiting-input handling, and prompt patterns for product marketing workflows.

The fastest install path for supported agents is:

```bash
npx skills install slideshot
```

For agents the installer does not detect, clone the repo and paste `slideshot/SKILL.md` into the agent's system prompt or rules.

## Troubleshooting

- **Agent does not support MCP and has no shell access.** Use the raw HTTP path. Slideshot's API is fully usable with `fetch` or any HTTP client.
- **Agent invents endpoint paths or option names.** Reload the skill into the agent's context.
- **MCP OAuth approval failed.** Re-run the approval flow; check that the agent's redirect URI is reachable.
- **Recording stops at a login page.** Save credentials for the target domain. See [Login to any web app](/docs/authentication).

## Related

- [Install skills](/docs/install-skills): how to install the Slideshot skill in any agent that loads instructions.
- [MCP versus CLI](/docs/mcp-vs-cli): when to use the MCP server vs the CLI.
- [API overview](/docs/api/overview): if your agent only speaks HTTP.