Agent Utility MCP

analyze_url_risk

Analyze URLs for structural and security risk signals before access.

inspect_file

Inspect file metadata and content signatures without executing the file.

inspect_command

Analyze shell commands for destructive, privileged, network or remote-execution behavior without executing them.

analyze_tool_call

Analyze proposed agent tool calls before execution and detect destructive actions, privilege changes, sensitive-data exposure, external transmission, financial operations, and other high-risk behavior.

Path scope is the signal

Damage is predicted by where an action lands, not what it's called. rm -rf ./dist scores 10. rm -rf /etc scores 90.

Policy without bypass

Callers pick permissive, balanced or strict. Critical rules ignore policy entirely — a compromised agent cannot disable the layer.

Read-only and private

Never executes anything. No external API calls. No payloads, paths or arguments in telemetry.

Lightweight protection before action.

Designed as a deterministic, read-only security preflight layer for AI agents. It executes no commands or files and makes no external API calls.

Deterministic analysisRead-onlyNo executionNo external APIs

MCP endpoint
https://agent-utility-mcp.insivotron.workers.dev/mcp

Status
Early preview · ruleset 1.0.0 · schema 0.3.0

Pricing
Free during product-market-fit validation