MCPVet
Is that MCP server
actually safe to install?
AI agents now install MCP servers, Claude Code skills, and plugins from strangers — with permission to run commands, read your secrets, and call the internet. MCPVet reads the source and tells you what it can really do, in seconds.
Try: @modelcontextprotocol/server-filesystem · github.com/modelcontextprotocol/servers
What MCPVet checks
Static analysis
Fetches the real source and flags shell execution, secret access, network calls, install hooks, and obfuscated payloads — with the exact file and line.
AI instruction review
Reads the skill text, manifests, and tool descriptions for hidden prompt-injection and data-exfiltration instructions that static rules miss.
Shareable verdict
Every scan gets a clean, linkable report card with a risk grade you can drop in a PR, a Discord, or your team's allowlist review.
Vet extensions inside Claude Code
Scan before you install — without leaving your agent. One step, free.
① MCP server
Adds a scan_extension tool your agent calls on its own before installing anything.
claude mcp add mcpvet -- npx -y github:LorenzoLombardi111/factory-mcpvet
Then just ask: “scan @some/mcp-server before I install it.”
② Skill
A lightweight skill — drop it in ~/.claude/skills/ and Claude vets any repo, package, or pasted SKILL.md on request.
npx -y degit github:LorenzoLombardi111/factory-mcpvet/skill/mcpvet ~/.claude/skills/mcpvet
No dependencies — calls the same hosted scanner over HTTPS.
Recently scanned
Stay ahead of risky extensions
Scanning is free forever. Pro & Team add monitoring, private scans, and policy.
Pro
- Private repos & skills
- Monitor your installed set
- Deeper analysis & history
Team
- Org allowlist & policy
- CI gate for unsafe extensions
- Continuous monitoring