policylayer

Scan your MCP config. See what your AI agent can do. Get a shareable report.

Quick start

npx -y policylayer

That's it. The CLI finds your MCP config, live-scans each server, classifies every tool, and prints a report URL.

What it does

1. Finds your config -- checks .mcp.json, ~/.claude.json, Claude Desktop, Cursor, VS Code, Windsurf, and Codex configs automatically
2. Live-scans servers -- spawns each MCP server, performs the JSON-RPC handshake, and discovers every tool with its full schema
3. Classifies tools -- checks against the PolicyLayer database of 2,500+ classified tools. Unknown tools are classified locally using schema analysis, blast radius detection, and verb matching
4. Generates a policy YAML -- suggested default rules for every tool, ready to use with Intercept
5. Prints a report URL -- permanent, shareable, no login required
6. Contributes back -- live-scanned tool data is contributed to the database so future scans are faster for everyone

Commands

npx -y policylayer

Options:

| Flag | Description | |------|-------------| | -d, --dir <path> | Directory to scan for config files (default: cwd) | | -o, --output <path> | Output path for policy YAML (default: policylayer.yaml) | | --no-live | Skip live scanning, classify from config only | | --no-report | Skip submitting report to PolicyLayer | | --timeout <ms> | Timeout per server in milliseconds (default: 30000) | | --json | Output results as JSON |

Examples

# Auto-detect config and scan
npx -y policylayer

# Scan a specific directory
npx -y policylayer -d ~/projects/my-app

# Skip live scanning (faster, less detailed)
npx -y policylayer --no-live

# Output as JSON for piping
npx -y policylayer --json

# Custom policy output path
npx -y policylayer -o my-policy.yaml

Config detection

The CLI searches these paths and uses all configs found:

| Client | Path | |--------|------| | Claude Code (project) | .mcp.json | | Claude Code (user) | ~/.claude.json | | Claude Desktop | ~/Library/Application Support/Claude/claude_desktop_config.json | | Cursor | .cursor/mcp.json | | VS Code | .vscode/settings.json | | Windsurf | ~/.codeium/windsurf/mcp_config.json | | Codex (project) | .codex/config.toml | | Codex (user) | ~/.codex/config.toml |

Privacy

The CLI never sends your raw config. Before anything leaves your machine:

• Environment variables are stripped
• Auth tokens are removed
• Only server names, package identifiers, and tool schemas are sent
• Live-scanned tool data (names, descriptions, schemas) is contributed to improve the database

Use --no-report to skip sending anything.

Registry lookups (stack, precheck, the PreToolUse hook) send identifier candidates only: npm package names, registry slug guesses, and config key names. Config contents, env values, command lines and file paths never leave the machine. If config key names are themselves sensitive, stack --packages-only restricts candidates to package identifiers. Lookups also maintain a local last-seen record cache at ~/.policylayer/precheck.json (record summaries only) that powers the "since last seen" diff.

Links

• Example report
• Scan your config online
• Policy library
• Intercept -- enforce limits on every MCP tool call