# agent-runtime-guard MCP server

> Provides fail-closed runtime interception for AI agents to block command injection, SSRF, and credential exfiltration before execution.

- Repository: https://github.com/DSHCorrectover/agent-runtime-guard
- Page: https://hostedmcp.org/servers/dshcorrectover/agent-runtime-guard
- Transport: stdio
- Stars: 2
- Latest version: 1.0.0

## Install

### Claude Code

```sh
claude mcp add agent-runtime-guard -- npx -y agent-runtime-guard
```

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

```json
{
  "mcpServers": {
    "agent-runtime-guard": {
      "command": "npx",
      "args": [
        "-y",
        "agent-runtime-guard"
      ]
    }
  }
}
```

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

```json
{
  "servers": {
    "agent-runtime-guard": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "agent-runtime-guard"
      ]
    }
  }
}
```

### VS Code CLI

```sh
code --add-mcp '{"name":"agent-runtime-guard","type":"stdio","command":"npx","args":["-y","agent-runtime-guard"]}'
```

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

```json
{
  "mcpServers": {
    "agent-runtime-guard": {
      "command": "npx",
      "args": [
        "-y",
        "agent-runtime-guard"
      ]
    }
  }
}
```

### Codex

```sh
codex mcp add agent-runtime-guard -- npx -y agent-runtime-guard
```

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

```toml
[mcp_servers.agent-runtime-guard]
command = "npx"
args = ["-y", "agent-runtime-guard"]
```

### Gemini CLI

```sh
gemini mcp add -s user agent-runtime-guard npx -- -y agent-runtime-guard
```

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

```json
{
  "mcpServers": {
    "agent-runtime-guard": {
      "command": "npx",
      "args": [
        "-y",
        "agent-runtime-guard"
      ]
    }
  }
}
```

## Tools (5)

- `verify_tool_call`
- `issue_evidence`
- `audit_mcp_config`
- `verify_intent_binding`
- `verify_receipt`

## FAQ

### What does agent-runtime-guard protect against?

It blocks destructive commands, SSRF (including cloud metadata), credential exfiltration, cross-tool attack chains, path traversal, and prompt injection.

### Does this require an API key or an account?

No, it does not require an API key or an account. The first run generates an Ed25519 keypair.

### Can the decision receipts be verified offline?

Yes, every decision emits Ed25519-signed evidence receipts that are fully verifiable offline.