start_vllm

THE RISK

High Risk

Why start_vllm needs a policy

This tool executes a complex external operation—launching a Docker/Podman container with vLLM—which constitutes code/service execution. While not directly destructive or financial, it triggers infrastructure-level changes and opens network services whose behavior and side effects depend on how the AI agent configures it (ports exposed, resource limits, model loaded, etc.).

From the tool's definition Tool description explicitly states it 'Start[s] a vLLM server in a Docker container' with 'Automatically detects platform...

Documented attack patterns abuse exactly the kind of access start_vllm gives an agent:

POLICY

How to control start_vllm

PolicyLayer is an MCP gateway — it sits between your AI agents and vLLM MCP Server, and nothing reaches the server without passing your rules. This is the rule we recommend for start_vllm:

policy.json

{
  "version": "1",
  "default": "deny",
  "tools": {
    "start_vllm": {
      "limits": [
        {
          "counter": "start_vllm_rate",
          "window": "minute",
          "max": 10,
          "scope": "grant"
        }
      ]
    }
  }
}

start_vllm stays usable, but rate-capped — a runaway agent can't fire it dozens of times a minute. Everything else on the server is denied unless you say otherwise.

Create a free account and register vLLM MCP Server — nothing to install.
Add this policy — paste it, or build it visually.
Point your MCP client (Claude, Cursor, anything) at your gateway URL.

RATE-LIMIT THIS TOOL →

Free to start. No card required.

EXPLORE

Related tools and policies

More vLLM MCP Server tools

Execute restart_vllm Restart a vLLM Docker container Execute run_benchmark Run a performance benchmark against the vLLM server using GuideLLM Execute stop_vllm Stop a running vLLM Docker container Execute vllm_chat Send a chat message to the vLLM server. Supports multi-turn conversations. Execute vllm_complete Generate text completion using vLLM. Good for code completion and text generation. Read get_model_info Get detailed information about a specific model Read get_platform_status Get platform information including Docker and GPU availability Read get_vllm_logs Get logs from a vLLM container to check loading progress or errors

All 12 vLLM MCP Server tools →

Execute tools on other servers

Android Forensics ADB MCP Server adb_shell_command EMBA-MCP run_emba_scan Frida Game Hacking MCP remove_breakpoint Procmon request_elevation

Go deeper

High-risk tools in vLLM MCP Server → Every high-risk tool this server exposes, in one view.
All high-risk MCP tools → Every high-risk tool across the catalogue, and the policy pattern that contains them.
The MCP Attack Database → Documented attack patterns against MCP deployments — and the policies that stop them.
Agent sandbox → Bounding what an agent can execute and where the boundary should sit.
Least privilege for MCP → Granting agents the minimum tool surface the task needs.
Rate Limiting MCP Tool Calls: A Practical Guide →
MCP Security: Why Prompt Guardrails Aren't Enough →

FAQ

Questions about start_vllm

What does the start_vllm tool do? +

Start a vLLM server in a Docker container. Automatically detects platform (Linux/macOS/Windows) and GPU availability. It is categorised as a Execute tool in the vLLM MCP Server MCP Server, which means it can trigger actions or run processes. Use rate limits and argument validation.

How do I enforce a policy on start_vllm? +

Register the vLLM MCP Server MCP server in PolicyLayer and add a rule for start_vllm: allow, deny, rate-limit, or require approval. Point your MCP client at the PolicyLayer proxy URL and the rule is enforced on every call, before it reaches vLLM MCP Server. Nothing to install.

What risk level is start_vllm? +

start_vllm is a Execute tool with high risk. Execute tools should be rate-limited and have argument validation enabled.

Can I rate-limit start_vllm? +

Yes. Add a rate_limit block to the start_vllm rule in your PolicyLayer policy. For example, setting max: 10 and window: 60 limits the tool to 10 calls per minute. Rate limits are tracked per agent session and reset automatically.

How do I block start_vllm completely? +

Set action: deny in the PolicyLayer policy for start_vllm. The AI agent will receive a policy violation error and cannot call the tool. You can also include a reason field to explain why the tool is blocked.

What MCP server provides start_vllm? +

start_vllm is provided by the vLLM MCP Server MCP server (micytao/vllm-mcp-server). PolicyLayer sits as a proxy in front of this server to enforce policies before tool calls reach the server.

Enforce policy on every vLLM MCP Server tool call.

Start from vLLM MCP Server, add the rest of your stack, and see everything your agents can call. Then put policy on all of it.

CHECK YOUR STACK →