execute_scratch · Jlab MCP: Risk & Policy

WHAT IT DOES

What execute_scratch does on Jlab

AI agents invoke execute_scratch to trigger actions in Jlab. What it does depends on the arguments the agent supplies, and its effects often reach beyond the immediate call — builds kicked off, notifications sent, workflows started.

WHY IT NEEDS A POLICY

Why execute_scratch needs a policy

This tool almost certainly executes arbitrary code (Python) on remote GPU-accelerated HPC nodes. Code execution is inherently an Execute category risk because the effects depend entirely on what code is run—from benign computation to data exfiltration, resource exhaustion, or lateral movement within a cluster.

From the tool's definition Tool name 'execute_scratch' in a server that 'enables LLMs to execute Python code on GPU-accelerated compute nodes within SLURM-managed HPC environments.' The sibling tool 'execute_code' and context of 'launching JupyterLab sessions' indicate code execution…

Attacks that exploit this kind of access

Recommended rule Ready

Verdict

execute_scratch Rate-limited

execute_scratch 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.

View as policy code

policy.json

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

RATE-LIMIT THIS TOOL →

Instant setup, no code required.

PolicyLayer is an MCP gateway: every tool call is checked against your rule before it reaches Jlab, so you can run agents against it safely.

FAQ

Questions about execute_scratch

What does the execute_scratch tool do? +

execute_scratch. It is categorised as a Execute tool in the Jlab MCP Server, which means it can trigger actions or run processes. Use rate limits and argument validation.

How do I enforce a policy on execute_scratch? +

Register the Jlab MCP server in PolicyLayer and add a rule for execute_scratch: 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 Jlab. Nothing to install.

What risk level is execute_scratch? +

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

Can I rate-limit execute_scratch? +

Yes. Add a rate_limit block to the execute_scratch 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 execute_scratch completely? +

Set action: deny in the PolicyLayer policy for execute_scratch. 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 execute_scratch? +

execute_scratch is provided by the Jlab MCP server (kdkyum/jlab-mcp). PolicyLayer sits as a proxy in front of this server to enforce policies before tool calls reach the server.