# falsepass MCP server

> Probes reinforcement learning environment verifiers to identify reward function defects like false passes and false fails.

- Repository: https://github.com/Super-Eric/falsepass
- Page: https://hostedmcp.org/servers/super-eric/falsepass
- Transport: stdio
- Stars: 1
- License: Apache-2.0
- Latest version: 0.15.0

## Install

### Claude Code

```sh
claude mcp add falsepass -- uvx falsepass
```

### Cursor (`~/.cursor/mcp.json`)

```json
{
  "mcpServers": {
    "falsepass": {
      "command": "uvx",
      "args": [
        "falsepass"
      ]
    }
  }
}
```

### VS Code (`.vscode/mcp.json`)

```json
{
  "servers": {
    "falsepass": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "falsepass"
      ]
    }
  }
}
```

### VS Code CLI

```sh
code --add-mcp '{"name":"falsepass","type":"stdio","command":"uvx","args":["falsepass"]}'
```

### Claude Desktop (`claude_desktop_config.json`)

```json
{
  "mcpServers": {
    "falsepass": {
      "command": "uvx",
      "args": [
        "falsepass"
      ]
    }
  }
}
```

### Codex

```sh
codex mcp add falsepass -- uvx falsepass
```

### Codex config.toml (`~/.codex/config.toml`)

```toml
[mcp_servers.falsepass]
command = "uvx"
args = ["falsepass"]
```

### Gemini CLI

```sh
gemini mcp add -s user falsepass uvx -- falsepass
```

### Gemini settings.json (`~/.gemini/settings.json`)

```json
{
  "mcpServers": {
    "falsepass": {
      "command": "uvx",
      "args": [
        "falsepass"
      ]
    }
  }
}
```

## Tools (1)

- `ratctl`

## FAQ

### What does falsepass do?

It probes RL environment verifiers for false passes (where bad completions are rewarded) and false fails (where good completions are not).

### Does falsepass require a model or GPU to run?

No, it treats the verifier as the system under test without needing a model, rollouts, or a GPU.

### Which frameworks are supported?

It works with verifiers, OpenEnv, TRL's GRPOTrainer reward functions, Inspect AI, and HUD.