> ## Documentation Index > Fetch the complete documentation index at: https://docs.firedrill.run/llms.txt > Use this file to discover all available pages before exploring further. # FAQ > Answers to common questions about running AI agent simulations with Firedrill. ## What is Firedrill? A simulation platform for AI agents. It runs your agent in a synthetic world of users, tools, data and scenarios, then evaluates how it handled each situation. Your code and model stay on your machine. ## What does practice mean here? Running your agent in controlled situations, again and again, and improving its prompt, policy or code from what you see. Firedrill does not retrain your model's weights. ## Who is Firedrill for? Teams building AI agents that talk to users or act in other services, such as support, sales, operations and browser agents. It is designed to be operated by coding agents like Claude Code, Codex and Cursor, with a person approving anything that runs code or uses credits. ## Does Firedrill upload my source code? No. Firedrill receives a descriptor with names and hashes, a digest of your local plan and what happens in each drill. Source code, file paths and secret values stay on your machine. ## Where does my agent run? On your machine. The CLI starts your agent's services as normal processes. They are not sandboxed, so only approve commands you trust. ## Does Firedrill pay for my agent's model calls? No. Your agent uses your own model keys, and your model provider bills you. Firedrill credits cover the simulated users, synthetic tools and scoring. ## Can my coding agent run drills through MCP? No. The MCP server cannot read your files or start processes. Your coding agent uses MCP to manage resources and read reports, and runs `firedrill run` through the CLI to run drills. ## Which operating systems are supported? You can sign in and set up from macOS, Linux and Windows. Running drills needs macOS or Linux on arm64 or x64. The CLI needs Node.js 22.23.4+ (22.x), 24.11.1+ (24.x) or 26.13.1+ (26.x). ## Do I need Docker? No. Your agent runs as normal processes on your machine. ## What kinds of agents can I simulate? Any agent you can reach in one of these ways: an HTTP endpoint, a command-line program over stdin and stdout, a function in your code, a web page through Playwright, or a webhook receiver. See [Drivers](https://docs.firedrill.run/connect/drivers.md). ## Which tools can my agent use during a simulation? Synthetic versions of services such as Gmail, Slack, Google Calendar, Stripe, HubSpot, Salesforce, GitHub, Linear, Jira and Notion. See [Synthetic tools](https://docs.firedrill.run/tools/overview.md). You can also build your own with a [custom tool](https://docs.firedrill.run/tools/custom-tools.md). ## Does a synthetic tool ever contact the real service? No. It keeps its own data. No email is sent, no payment is made and no real record changes. ## How is a drill scored? Each evaluation turns your criteria into a rubric and rates the evidence on five levels, which become a score of 0, 25, 50, 75 or 100. The evaluation is met when the score reaches your threshold. Exact checks count things like tool calls and pass or fail without a model. See [Evaluations](https://docs.firedrill.run/build/evaluations.md). ## What does "not assessable" mean? The evidence an evaluation needs was not recorded, or that kind of evidence cannot be scored yet. It is not a zero and not a failure of your agent. ## Can I run drills in parallel? Yes, when your agent's state can be reset for each drill. Set every state domain to `independent`, set the world's `reset` to `per_drill` and raise `max_parallel`. ## Can I rerun a drill? Yes. To test a change to your agent, plan and apply the same world again to get a new campaign and compare the reports. To score an existing drill again without running your agent, [reassess](https://docs.firedrill.run/results/reassess.md) it. ## How do I connect Firedrill to Claude Code? ```bash theme={null} claude mcp add --transport http firedrill https://mcp.firedrill.run ``` Then run `/mcp` in Claude Code and sign in. See [Set up your coding agent](https://docs.firedrill.run/agents/setup.md). ## How do I connect Firedrill to Codex? ```bash theme={null} codex mcp add firedrill --url https://mcp.firedrill.run codex mcp login firedrill ``` ## Is there an API key? No. The CLI and MCP clients sign in through your browser. The CLI keeps your sign-in in your operating system's credential store. > [card] [Quickstart](https://docs.firedrill.run/quickstart.md) > > Run your first drill. > [card] [Troubleshooting](https://docs.firedrill.run/reference/troubleshooting.md) > > Error codes and fixes. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.