Heads up This site is currently under heavy development.
← all tools
◆ AI/LLM Security

Snyk Agent Scan

open-source

Summary

Snyk Agent Scan is an open-source command-line tool that discovers and scans agent components on a machine — MCP servers, skills, and harnesses — for prompt injections, tool poisoning, and other vulnerabilities. It is distributed via PyPI and free to use, run either as a one-off scan across all known MCP configs and skills or targeted at a single config or skill file, with a CI mode that auto-starts MCP servers for automated checks. It's aimed at teams building or operating AI agents who need to vet third-party skills and MCP integrations before trusting them, rather than at traditional application security scanning. Its README frames it around the emerging agent-skill ecosystem rather than positioning against a named competitor, so it sits in the newer category of AI agent and MCP security tooling.

What Snyk Agent Scan answers

Can I trust the CLI output to stay stable across versions?

No — issue codes, severity labels, and response fields are explicitly experimental and can change without notice, so scripting around specific output values is not recommended.

Is the risk scoring itself stable, even if the output format isn't?

Yes for enterprise use — the underlying discovery and risk assessment feeding the Evo platform is stable and supported, even though the raw CLI representation of it is not.

Do I need to configure anything before it can see what's on the machine?

No — it discovers MCP configs and skills on its own, though it can also be pointed at a single file if you want to check one thing without a full scan.

Can I see what it found without waiting on a security verdict?

Yes — it can list every discovered MCP entity and skill file on its own, separately from running the vulnerability analysis on them.

Does using this commit me to sending data anywhere?

No — results display locally by default, and pushing results to a remote analysis server is a separate, explicit step requiring its own authentication.

Can it tell which machine a set of results came from in a shared or fleet setting?

Yes — a machine identifier such as a hostname or serial number can be attached to a scan for enterprise deployments tracking multiple hosts.

Release history

Nothing shipped since we started watching. Releases and docs updates land here as they happen.

last release Aug 28, 2026 · watching since Jul 18, 2026

my-toolchain — 0 tools
paste an install list to detect your tools

A brew list, a Brewfile, requirements.txt, a Dockerfile — or just the product names, free-form. Nothing leaves your browser.

    browse all tools →