@vantinel/mcp

MCP server that exposes your Vantinel AI agent monitoring data as tools for Claude Code, Claude Desktop, and any MCP-compatible client.

Ask your AI assistant questions like:

• "Are any of my agents in a zombie loop right now?"
• "How much has session sess_abc123 spent so far?"
• "Block the execute_code tool for project X"
• "Show me the last 10 anomalies"

Installation

npm install -g @vantinel/mcp

Setup

Claude Code

Add to your Claude Code MCP settings (~/.claude/settings.json or via /mcp):

{
  "mcpServers": {
    "vantinel": {
      "command": "npx",
      "args": ["-y", "@vantinel/mcp"],
      "env": {
        "VANTINEL_API_KEY": "your-api-key"
      }
    }
  }
}

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "vantinel": {
      "command": "npx",
      "args": ["-y", "@vantinel/mcp"],
      "env": {
        "VANTINEL_API_KEY": "your-api-key",
        "VANTINEL_BASE_URL": "https://app.vantinel.com"
      }
    }
  }
}

Environment Variables

| Variable | Required | Description | |---|---|---| | VANTINEL_API_KEY | Yes | Your Vantinel API key (from dashboard → Settings → API Keys) | | VANTINEL_BASE_URL | No | Defaults to https://app.vantinel.com |

Available Tools

Observability

| Tool | Description | |---|---| | list_projects | List all projects in your Vantinel organization | | get_sessions | Get active and recent agent sessions, filterable by project and time range | | get_session_detail | Deep-dive into a single session: cost, call count, decisions | | get_metrics | Aggregated metrics: tool call rates, costs, latency percentiles | | get_alerts | Recent alerts: zombie loops, budget warnings, anomalies, blocked tools | | get_anomalies | Anomaly detection feed: latency spikes, frequency surges, cost outliers | | get_tool_errors | Tool failure events captured via captureError() — rates, types, retry patterns | | check_connectivity | Verify your Vantinel collector is reachable from the MCP server |

Guardrails

| Tool | Description | |---|---| | get_approvals | List tool calls awaiting human approval | | approve_tool_call | Approve or deny a pending tool call with optional reason | | get_policy | View current policy: budget caps, blocked tools, approval rules | | update_policy | Update policy: set budget caps, block tools, require approval for patterns |

Budget Intelligence

| Tool | Description | |---|---| | get_forecast | Predict when a session/project will exhaust its budget based on current burn rate — alerts you before budget runs out | | get_shadow_blocks | Tool calls Vantinel would have blocked in shadow mode — use to prove value before enabling enforcement |

Example Usage

Once configured, you can ask Claude natural-language questions:

"Check if any sessions are in a zombie loop right now"

"What's the current burn rate for project X? Will it exceed budget?"

"Block the execute_code tool for my production project"

"Approve the pending delete_users tool call — it was authorized by the user"

"Show me all anomalies from the last 24 hours"

License

MIT © Vantinel AI