# Sentinel Scan MCP server

> Statically scans MCP tool manifests and runs prompt-injection attack suites to detect security vulnerabilities like tool poisoning and jailbreaks.

- Repository: https://github.com/Ventrova/sentinel-scan-cli
- Page: https://hostedmcp.org/servers/ventrova/sentinel-scan-cli
- Transport: stdio
- Stars: 3
- License: MIT
- Latest version: 1.4.16

## Install

### Claude Code

```sh
claude mcp add sentinel-scan-cli -- npx -y sentinel-scan-cli
```

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

```json
{
  "mcpServers": {
    "sentinel-scan-cli": {
      "command": "npx",
      "args": [
        "-y",
        "sentinel-scan-cli"
      ]
    }
  }
}
```

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

```json
{
  "servers": {
    "sentinel-scan-cli": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "sentinel-scan-cli"
      ]
    }
  }
}
```

### VS Code CLI

```sh
code --add-mcp '{"name":"sentinel-scan-cli","type":"stdio","command":"npx","args":["-y","sentinel-scan-cli"]}'
```

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

```json
{
  "mcpServers": {
    "sentinel-scan-cli": {
      "command": "npx",
      "args": [
        "-y",
        "sentinel-scan-cli"
      ]
    }
  }
}
```

### Codex

```sh
codex mcp add sentinel-scan-cli -- npx -y sentinel-scan-cli
```

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

```toml
[mcp_servers.sentinel-scan-cli]
command = "npx"
args = ["-y", "sentinel-scan-cli"]
```

### Gemini CLI

```sh
gemini mcp add -s user sentinel-scan-cli npx -- -y sentinel-scan-cli
```

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

```json
{
  "mcpServers": {
    "sentinel-scan-cli": {
      "command": "npx",
      "args": [
        "-y",
        "sentinel-scan-cli"
      ]
    }
  }
}
```

## Tools (10)

- `tool_description_injection`
- `tool_name_shadowing`
- `excessive_agency_schema`
- `indirect_injection_surface`
- `unpinned_remote_source`
- `hardcoded_credential`
- `overbroad_tool_scope`
- `missing_provenance`
- `missing_hitl_confirmation`
- `hidden_unicode_instructions`

## FAQ

### Can I use it against my own LLM endpoint?

Yes, it works against any endpoint that uses the OpenAI-compatible chat completions format, such as OpenAI, Azure, Ollama, vLLM, and LM Studio.
