AI assistants
Let Claude Code, Codex, Cursor and other assistants build worlds, run evaluations and explain results for you. What the plugin installs, and what an assistant may do.
An AI assistant can drive Rehearsal for you: add an application, build a world, run an evaluation, read why an episode failed, and propose a better agent. You install one plugin and sign in once.
What the plugin installs
| Part | What it does |
|---|---|
| The Rehearsal skill | Instructions for the assistant: the workflows, how to read results, and the rule to state the cost and ask before it spends money |
| The Rehearsal MCP server | 18 tools that the assistant calls. It runs on your machine from the rehearsal-kit package, started with uv |
The MCP server holds no data. Each tool calls the Rehearsal server with your saved sign-in.
Set up
Install uv
The assistant starts the MCP server with uvx, which is part of uv. Install uv
first.
Install the plugin for your client
For another client, see Other clients below.
Sign in once
Every client uses the same sign-in. Run it in a terminal:
uvx --from rehearsal-kit rehearsal loginAdd --url <address> for a server other than the default. Your browser opens the console. Check that the code
matches your terminal, then choose Approve. The terminal receives a key with Build and evaluate access,
saved for your user only.
On a machine with no browser, use rehearsal login --with-key and paste a key from the console's API keys
page.
Check that it works
Ask the assistant to show your Rehearsal workspace status. It calls workspace_status and reports your plan and
what exists in the workspace.
Without a saved sign-in the MCP server still starts. Its tools then tell the assistant to ask you to run
rehearsal login.
What an assistant may do
- It has the access of your key, and no more. A key with "Build and evaluate" access can add applications, build worlds, run evaluations and read results. It cannot manage keys or billing.
- It asks before it spends. The skill tells the assistant to state the cost and wait for your yes before each of the four tools that spend model budget.
- Your plan is a hard stop. The plan's limits and the monthly model allowance apply to every request, whatever the assistant decides.
- You can stop it at once. Revoke the key on the API keys page. The assistant loses access immediately.
| The assistant must ask before | Typical cost | Typical time |
|---|---|---|
build_world | $3 to $8 | 1 to 3 hours |
run_evaluation | about $0.02 for each episode | 2 to 10 minutes |
improve_agent | under $0.10 | about 1 minute |
recheck_world | about $0.03 for each job | 5 to 20 minutes |
Never paste a key into the chat
The assistant does not need to see your key, and the skill tells it never to ask for one. If it asks, run
rehearsal login in a terminal instead.
What to ask
Show my Rehearsal workspace status.Evaluate my support agent on the published Gitea world, on the dev jobs, with two repeats.Why did the failed episodes of that run fail? Group them by cause.Improve the agent from those failures. Then compare the old and the new version on holdout.Act as the agent yourself on one dev job, so that I can see how a careful agent does it.Other clients
The plugin follows the Agent Plugins layout, and the MCP server is a standard stdio server. Any client that supports one of them can use Rehearsal.
Clients with a plugin system
| Client | Install |
|---|---|
| GitHub Copilot CLI | copilot plugin marketplace add MitudruDutta/rehearsal, then copilot plugin install rehearsal@rehearsal |
| Factory Droid | droid plugin marketplace add MitudruDutta/rehearsal, then droid plugin install rehearsal@rehearsal |
| Augment (Auggie) | auggie plugin marketplace add MitudruDutta/rehearsal, then auggie plugin install rehearsal@rehearsal |
| Grok Build | grok plugin marketplace add MitudruDutta/rehearsal, then grok plugin install rehearsal --trust |
| Antigravity CLI | Clone the repository, then agy plugin install ~/.rehearsal-src/agent-plugin/providers/antigravity |
| Hermes Agent, OpenClaw, OpenHands, NanoClaw | Point the client at the agent-plugin/ folder in a clone of the repository |
Any client with MCP
Add the MCP server to the client's MCP settings. Most clients accept this entry:
{
"mcpServers": {
"rehearsal": {
"command": "uvx",
"args": ["--from", "rehearsal-kit[mcp]>=0.1.1", "rehearsal", "mcp"]
}
}
}Then install the skill into the client's skills folder, if the client reads skill folders:
uvx --from rehearsal-kit rehearsal skill install --dir <skills_folder>The place of the MCP settings differs by client. The command and its arguments are the same everywhere.
An assistant with a shell and no MCP
The skill alone is enough: it falls back to the rehearsal command. Install the skill, then install the package
and sign in:
pip install rehearsal-kit
rehearsal loginNext
Was this page helpful?