Install

Three ways. Pick yours.

Hosted on GitHub. Signed releases. Reproducible builds. The same binary that runs in our SaaS — no special sauce.

▴

Package managers

macOS · Linux · Windows

$ brew install mcpscanner/tap/mcpscanner

> scoop install mcpscanner

Homebrew for macOS & Linux
Scoop for Windows (scoop bucket add mcpscanner …)
Auto-updates on every release

recommended for CI

⤓

Binary download

all platforms

$ curl -fsSL https://install.mcpscanner.dev/install.sh | sh

Single static binary, no runtime
Pin a version: VERSION=1.0.1
SHA-256 checksums + cosign signatures
GitHub Releases: every tag, every platform

Browse releases →

◇

From source

Go 1.24+

$ git clone https://github.com/mcpscanner/cli

Apache 2.0 — fork it, audit it, ship it
Or one-liner: go install github.com/mcpscanner/cli@latest
Build / test instructions in the README

View on GitHub →

Usage

Built for CI. Friendly for humans.

JSON / SARIF output, configurable severity thresholds, exit codes for build gates. Use it however your pipeline wants.

~ mcpscanner v1.0.1

# Quick scan

$ mcpscanner scan https://mcp.example.com/mcp

▸ resolving https://mcp.example.com/mcp

✓ JSON-RPC 2.0 handshake

✓ 8 tools enumerated

✗ 3 critical · 4 high · 3 medium · 2 low

score: 45 / 100 · grade: F

# JSON output for CI

$ mcpscanner scan https://mcp.example.com/mcp --format json --output report.json

→ 12 findings written to report.json

# Standalone HTML report

$ mcpscanner scan https://mcp.example.com/mcp --format html --output report.html

# With auth token + interactive mode

$ mcpscanner scan --interactive --token "sk-your-token"

# Scan a Claude Desktop / Cursor config file

$ mcpscanner scan --config ~/.claude/claude_desktop_config.json

▸ parsing 3 MCP servers from config

✗ hardcoded secret OPENAI_API_KEY · unpinned npx version

# Internal IPs (disabled by default for safety)

$ mcpscanner scan http://192.168.1.100:8080/mcp --allow-internal

$ ▌

Same engine as the SaaS

Clean-room Go port of the scanner engine. Identical findings, identical scoring.

Three output formats

--format text for humans, json for CI, html for a standalone dark-themed report.

Self-protecting

SSRF guard, redirect block, 10 MB response cap, 500-tool cap. Refuses internal IPs unless you opt in.

Apache 2.0 source

Public on github.com/mcpscanner/cli. Fork it, audit it, contribute.

Reference

Output formats & flags.

Three formats, eight flags. Full reference in the README — the basics live here.

Output formats

Format	Flag	What it's for
Text	`--format text`	Human-readable terminal output (default)
JSON	`--format json`	Structured payload for CI / SIEM ingestion
HTML	`--format html`	Standalone dark-themed report you can email

Flags

Flag	Short	Default	Description
`--format`	`-f`	text	Output format
`--config`	`-c`	—	Scan a Claude Desktop / Cursor config file
`--output`	`-o`	stdout	Write report to a file
`--timeout`	`-t`	60	HTTP timeout in seconds
`--token`	—	—	Auth token, sent as `Bearer`
`--interactive`	`-i`	false	Prompt-driven scan flow
`--no-progress`	—	false	Disable live progress display
`--allow-internal`	—	false	Allow scanning private / loopback IPs

Open source & Apache 2.0. If the CLI helps you, a GitHub star goes a long way. ★ Star on GitHub

mcpscanner, in your terminal.