agentgate scan

Scan MCP servers for security issues.

agentgate scan [target] [options]

Without a target, AgentGate auto-discovers MCP client configs (Claude Desktop, Claude Code, Cursor, VS Code, Codex, OpenCode). Pass a directory to also run a repo/source scan over it, or a config file to scan just that config.

Modes

Mode	What it does
static (default)	Analyzes client configs and (for a directory target) source files. Never executes server code.
`--live`	Additionally connects to each stdio server, performs the MCP handshake, and analyzes the tool surface the server actually exposes. Explicit opt-in because it runs server code.

Options

Flag	Default	Description
`--live`	off	Connect to stdio servers and analyze their live tool surface.
`-c, --config <file>`	auto-discover	Explicit MCP client config file (skips auto-discovery). Codex `config.toml` and OpenCode `opencode.json` are also understood.
`-s, --server <names...>`	all	Restrict to specific server names.
`-f, --format <format>`	`table`	Output format: `table`, `json`, `sarif`. JSON follows the scan output spec.
`-o, --output <file>`	stdout	Write the report to a file.
`--fail-on <severity>`	off	Exit non-zero when findings reach this severity: `info`, `low`, `medium`, `high`, `critical`.
`-t, --timeout <ms>`	`15000`	Per-server connect timeout for `--live`.

Examples

agentgate scan                                  # audit everything your clients are configured to run
agentgate scan --live                           # also audit the live tool surface
agentgate scan --format json -o report.json     # machine-readable report (open it in the report viewer)
agentgate scan --format sarif -o report.sarif   # for GitHub code scanning
agentgate scan path/to/repo                     # source-level scan of an MCP server repo
agentgate scan -c ~/.cursor/mcp.json --fail-on high

Rules

Findings come from seven rules, one per category — see the rule reference for what each detects: tool-poisoning, credential-leak, overprivileged, auth-missing, ssrf, rce-vectors, supply-chain.

Exit codes

Code	Meaning
0	No findings at or above the `--fail-on` threshold (or no `--fail-on` given).
1	Findings at or above the threshold.
2	Execution error (bad target, unreadable config).