# Firedrill agent quickstart

Firedrill is a simulation platform for AI agents. Connect your coding agent to
prepare and operate simulations of the agent you're building. Your target code,
model configuration and runner stay in your repository or CI.

## Remote MCP

Server URL: `https://mcp.firedrill.run/mcp`

Use Streamable HTTP and your client's browser authorization. Remote MCP does not
require installing a local Firedrill MCP package or the Firedrill CLI.

The [agent quickstart](https://firedrill.run/agents) includes installation buttons
for Cursor, VS Code, and VS Code Insiders. Review the connection in your client
and authorize Firedrill in the browser. Installation links do not complete
authorization for you. For Claude, open
[connector settings](https://claude.ai/customize/connectors), add a custom
connector named Firedrill with the server URL above, and connect it. Availability
depends on your Claude account; Claude Code uses its own command below.

For Codex:

```sh
codex mcp add firedrill --url https://mcp.firedrill.run/mcp
codex mcp login firedrill
```

For Claude Code:

```sh
claude mcp add --scope project --transport http firedrill https://mcp.firedrill.run/mcp
```

Then open `/mcp` in Claude Code and authorize Firedrill in the browser.

For Cursor, merge the following into `.cursor/mcp.json`. For Kimi Code, merge it
into `.kimi-code/mcp.json`. Preserve all existing servers:

```json
{
  "mcpServers": {
    "firedrill": {
      "url": "https://mcp.firedrill.run/mcp"
    }
  }
}
```

Authorize in Cursor's MCP settings. In Kimi Code, run
`/mcp-config login firedrill`, then start a new session.

For VS Code/Copilot manual setup, merge this into `.vscode/mcp.json`, preserving
existing servers. This is separate from the Codex extension's MCP settings:

```json
{
  "servers": {
    "firedrill": {
      "type": "http",
      "url": "https://mcp.firedrill.run/mcp"
    }
  }
}
```

Start the server in VS Code's MCP controls and complete its browser sign-in.

## CLI

For execution in your repository or CI, start with Node.js 22 or later:

```sh
npx firedrill-run init
```

The optional wizard offers project or global CLI installation, MCP configuration,
skills, and a TypeScript or Python SDK. Use arrows to choose, Space to toggle
checkboxes, and Enter to continue. Review the changes before confirming, or skip
every addition. It then offers browser authorization and project selection;
MCP authorization remains separate. Use `init --skip-install` for account/project
setup without optional installs. CI never prompts or installs implicitly.

## Start with your task

Tell your coding agent: “Use Firedrill to simulate this agent. Start with one useful
scenario, show me what happened, and save the report.”

The agent should inspect your real entry point, choose relevant synthetic Tools
and data when needed, run within your authority, and report actual outcomes and
available evidence. Assessment is optional. Different valid outputs need not be
failures; use explicit checks or a quality rubric appropriate to your goal.

Firedrill is free for now. Your model-provider and runner costs remain separate.
Do not send secrets in prompts or silently connect simulated workflows to real
production services.

[Agent setup guide](https://firedrill.run/index.md) ·
[Operator guide](https://firedrill.run/start.md) ·
[Documentation](https://docs.firedrill.run/) ·
[Open your reports](https://app.firedrill.run/)