@mdevs/mai-cli@1.0.18
The mAI agent, directly in your terminal!
AI Security Review
scanned 1mo ago · by lpm-firewall-aiThis is a Claude Code-style AI CLI whose automatic postinstall registers a first-party Chrome MCP native host, which is guarded platform extension setup rather than a concrete malicious chain. Scanner similarity labels on dist chunks are noisy leftovers of that fork and do not show hidden exfiltration or foreign-agent hijack in the shipped install scripts.
Decision evidence
public snapshot- package.json postinstall automatically runs both the ripgrep vendor script and Chrome MCP setup.
- Unless CLAUDE_CODE_SKIP_CHROME_MCP_SETUP=1, setup-chrome-mcp.mjs runs fix-permissions, register --browser chrome, and doctor via @claude-code-best/mcp-chrome-bridge.
- Bundled Chrome setup writes a native host manifest named com.anthropic.claude_code_browser_extension.json when that feature is enabled.
- This package is an AI terminal CLI; Chrome MCP and ripgrep match that product.
- scripts/postinstall.cjs only vendors microsoft/ripgrep-prebuilt into this package tree and treats download failure as non-fatal.
- mai.val.run login and API-key storage run only after the user starts mai and logs in.
- There is no runtime self-dependency, and the shipped install scripts do not write Cursor, Copilot, or other foreign agent settings files.
- Scanner malware-similarity hits are Claude Code fork chunks such as SSH, REPL, and computer-use, not a hidden install payload.
Source & flagged code
26 flagged · loading sourcePackage defines install-time lifecycle scripts.
package.jsonView on unpkgInstall-time lifecycle script is not statically allowlisted and needs review.
package.jsonView on unpkgFindings
Affected versions and remediation
This report applies to @mdevs/mai-cli@1.0.18.
See all versions of @mdevs/mai-cli for other recorded verdicts.
- Review the evidence and your use of @mdevs/mai-cli@1.0.18 before allowing it. Restrict the permissions described in this report.
- Choose an independently verified alternative or release. This report does not establish that other versions are safe.
Evidence last updated: .
Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.