Skip to main content
Firedrill is a simulation platform for AI agents. Your coding agent operates it from the application’s repository: it inspects the actual application, prepares a bounded scenario set, connects the real entry point, and returns retained evidence. The AI agent under simulation is separate from the coding agent doing this work.

​
Choose your host

Connect your coding agent directly to Firedrill’s remote MCP at https://mcp.firedrill.run/mcp. No local MCP package, Node.js installation, or Firedrill CLI is required for this connection. The skills plugin supplies the workflow instructions separately.
Add to CursorAdd to VS CodeVS Code Insiders
Open the installed desktop client, review the server configuration, then complete its Firedrill browser sign-in. VS Code connects through Copilot; the Codex extension uses Codex’s separate MCP settings. These buttons do not install skills, authorize access, or run a simulation. For Codex, Claude Code, or Kimi Code, follow the native setup below. Already connected? Keep that connection and inspect its available tools instead of adding another entry.
Approve the Firedrill sign-in in your browser. Native MCP registration and plugin installation are user-wide, across projects. Start a new session in the application repository, inspect /mcp, and select the Firedrill plugin skill. If your version lacks codex plugin add, install Firedrill from the added marketplace using /plugins, or use the portable skills alternative below. No trust or tool approval is implied. See the official Codex MCP guide and plugin guide.
The plugin supplies the firedrill simulation skill and firedrill-service-authoring skill. Adding a server or installing a plugin does not authorize Firedrill access: complete the browser sign-in separately. Neither step runs your agent, approves tools, or uploads your source. The native host owns configuration, installation scope and removal. Do not overwrite an existing Firedrill entry or install duplicate plugin and direct-skill copies. Follow the WorkOS browser prompts using the intended Firedrill account and organization. Your host stores the OAuth credential; do not paste tokens into chat or commit them to configuration. This authorization is for Firedrill, not your coding-agent provider or the application being simulated.

​
Start with a bounded request

Ask your agent:
Use Firedrill to simulate this application’s real entry point. Inspect its invocation, completion, dependencies, and reset boundary first. Propose a small scenario set and explain how many actual executions it needs. Run it after my approval, then return the report link, observed behavior, and any missing evidence. Do not add grading unless it helps answer my question.
The agent should use an existing command or HTTP path when suitable, or a native Python/TypeScript callback. It must not replace the application with its own generated response. New synthetic Tools are needed only when the question requires controlled dependencies. The remote MCP can prepare simulations, manage synthetic Tools, and inspect saved results. It cannot execute a process on your laptop. When actual execution is needed, your coding agent runs the CLI or SDK in your application repository.

​
Add a local execution client

For command or HTTP simulations, initialize the CLI with Node.js 22 or newer:
The optional setup wizard offers project or global CLI installation, MCP configuration, skills, and a TypeScript or Python SDK. Use arrow keys for a single choice, Space for checkboxes, and Enter to continue. Skip any or every addition; review the exact changes before confirming. No SDK is required for CLI/MCP use. init then offers browser sign-in when needed and selects or creates the project and its Default environment. Use the same organization and project selected by your MCP connection. CLI sign-in is separate from the MCP host’s OAuth session; do not copy credentials between them. Tools are optional. No global installation is required. Use npx firedrill-run for later commands, or choose a global installation to use firedrill. Reopen optional choices with npx firedrill-run install; skip them with npx firedrill-run init --skip-install. See CLI setup. For a callable, use the TypeScript or Python SDK. CI uses an authorized scoped machine credential from its secret store and never prompts for browser approval.

​
Skills-only installation

If you already have your remote MCP connection, use the portable skill installer instead of the native plugin—not in addition to it: Requires Node.js 22.20 or newer.
Choose the host supported by your skill installer. The public skills release contains the reusable instructions; skills.sh provides a discovery listing. Installing instructions alone does not install an execution client or authorize an account. Use one skill installation path per host/project.

​
Inspect, update, or remove setup

For native installation, use your host’s plugin and MCP managers. Removing a plugin does not necessarily remove a separately registered MCP server. Removing an integration does not revoke Firedrill account access or delete saved results. Use your host’s MCP logout action to clear its stored authorization, and its server manager to remove the remote entry. CLI sign-in and CI credentials are separate: firedrill logout ends CLI sign-in, while machine credentials must be revoked by their exact ID. None of these actions deletes retained reports. See operator MCP for permissions, queueing, and retained inspection. The agent setup guide gives coding agents a copyable first-trial workflow, scope checks, and recovery steps.