Install · Tools · Agents · When to use · Security · zfuzz.com

Zfuzz — security for everyone who builds with AI

You ask. Your AI answers with real scanners — not guesses.

Catch SQL injection, leaked API keys, and vulnerable dependencies while you build — no security background required. You don't run tools. You don't read dashboards. You just talk to your AI, and Zfuzz gives it real answers.

$0 · Apache-2.0 · No account · No API key · 100% local · Nothing leaves your machine

Made for vibe coders and developers alike. It plugs straight into the AI tools you already use — Cursor, Claude Code, Codex, Gemini CLI, OpenCode — and the web builders AI Studio, v0, and Lovable.

⚡ Add it in 10 seconds

Pick your tool. Copy one line. Done. No setup wizard, no account, no config to learn.

Cursor — one click

Click the button → Cursor opens → press Install. That's your two clicks.

Claude Code

Paste this in the chat and hit enter:

claude mcp add zfuzz -- npx -y @zfuzz/mcp

VS Code (Copilot)

code --add-mcp '{"name":"zfuzz","command":"npx","args":["-y","@zfuzz/mcp"]}'

Codex

codex mcp add zfuzz npx -y @zfuzz/mcp

Gemini CLI

gemini mcp add zfuzz npx -y @zfuzz/mcp

Lovable · v0 · AI Studio — in your browser, no install

Open your tool's MCP / Integrations settings and paste this URL:

https://api.zfuzz.com/mcp

Claude Desktop — Settings → Developer → Edit Config, then add:

{
  "mcpServers": {
    "zfuzz": { "command": "npx", "args": ["-y", "@zfuzz/mcp"] }
  }
}

OpenCode — add to opencode.json:

{
  "mcp": {
    "zfuzz": { "type": "local", "command": ["npx", "-y", "@zfuzz/mcp"] }
  }
}

Self-host the web (HTTP) endpoint instead of using the hosted URL:

npx -y @zfuzz/mcp --transport http --port 8099

Then point your tool at http://localhost:8099/mcp. (Use port 8099, not 8090 — 8090 is the Zfuzz platform's own port.)

That's it. Keep talking to your AI like normal — the first time your code needs a security check, it calls Zfuzz on its own.

Related MCP server: AgentShield

Agent compatibility matrix

Works with every major AI coding agent — and any MCP-compatible client. One line locally, or a hosted URL in the browser. Same 10 tools everywhere.

No per-agent config to learn — the exact copy-paste line for each is under Add it in 10 seconds.

What it feels like

You: Is my login page safe?

AI: [scans your file] Found 2 issues:

1. SQL injection on line 42 — anyone could read your whole database.

2. No rate limiting on the login — bots can guess passwords forever. Here's the fix… [writes the patch]

You: Did I leave any passwords or keys in my code?

AI: [scans the project] Yes — your OpenAI key is hard-coded in config.js. Anyone who sees this repo can use it on your bill. Move it to an environment variable — here's how…

You: Are the packages I installed safe?

AI: [checks your dependencies] 2 have known security holes. lodash lets attackers run code on your server. One command fixes both: npm update lodash axios.

You never typed a command, opened a scanner, or read a report. You just asked.

See it in action

You vibecode it. It ships with holes you can't see. Zfuzz catches them all — before you ship, for $0.

You don't need to understand security

• You never run a scanner yourself — your AI does, automatically, when it matters.

• You never read a dashboard — answers come back in plain English, in your chat.

• You never pay and nothing leaves your computer — no account, no cloud, no API key.

If you can copy-paste one line, you're covered.

What's under the hood

10 real tools your AI can call (it picks the right one — you don't have to):

Built in Rust for sub-second answers. The AI brain is your editor's own model (Claude/GPT) — Zfuzz adds the security muscle, not another subscription.

What's inside

What a scan actually returns

You: Is this skill safe to install?

AI: [runs scan_skill]

Verdict: DO NOT INSTALL   (risk 100/100, 6 findings)
 CRIT  ZF-C001   Prompt Injection Pattern Detected         @ description
 MED   ZF-SK001  Overly Broad Skill Trigger                @ description
 HIGH  ZF-LP001  Wildcard Permission Grant (Bash(*), *)    @ allowed-tools
 CRIT  ZF-C002   Pipe-to-Shell Download-Exec (curl | bash) @ setup.sh
 HIGH  ZF-H001   Hardcoded Secret: AWS Access Key ID       @ setup.sh

This skill hides "ignore all previous instructions" in its description, grants itself every tool, and its setup script pipes a remote payload straight into a shell. Don't install it.

Every finding is from a real deterministic scanner — there is no model guessing in the scan path.

Why not Snyk or Semgrep?

When to use · When to skip

Great fit if you…

• Vibecode apps with AI (Lovable, Cursor, v0, Bolt, Claude Code) and want them checked before you ship — without learning security.

• Want leaked keys, injectable code, and risky dependencies caught right in your editor, in plain English.

• Want to vet an MCP server or agent skill before you install it.

Skip it if you…

• Already run a full security team + pipeline and don't want findings in the editor.

• Work fully offline with no Node.js (the scanner binary is pulled via npx / @zfuzz/cli).

• Need a hosted dashboard with audit logs and SSO — that's the Zfuzz platform, not the free MCP.

Updating

npx -y @zfuzz/mcp always resolves the latest published version — most people never do anything.

npm view @zfuzz/mcp version      # see the latest release
npm i -g @zfuzz/cli@latest       # only if you pinned the CLI globally

Pinned a version in your MCP config? Bump it (or drop the pin) and restart your agent.

Documentation

Free · Open · Local

No API keys. No cloud account. No telemetry. Runs 100% on your machine — your code never leaves it. Apache-2.0 licensed, open source.

Contributing

Issues and PRs welcome — see CONTRIBUTING.md for dev setup. Found a vulnerability? Please follow SECURITY.md — don't open a public issue.

Community & support

• Questions / bugs — open an issue on this repo.

• Security reports — see SECURITY.md.

• The platform — zfuzz.com.

License

Apache-2.0 — free & open source. © Zfuzz

Part of the Zfuzz security platform.