MCP Server Risk Checker
Paste an MCP server config and scan it for risk signals before you install it.
About MCP Server Risk Checker
MCP servers run with real access to your filesystem, shell, and network, but the ecosystem has no package-manager-style review step — you install a stranger's server by pasting their config into claude_desktop_config.json (or Claude Code's MCP settings) and trusting it. Reports from 2026 security audits found a large share of public MCP servers vulnerable to command injection or path traversal, and a supply-chain vulnerability in the MCP SDKs itself affected millions of downloads. This tool runs a set of heuristic checks against your pasted config — shell interpreters invoked directly, broad filesystem paths, unpinned package versions, non-HTTPS remote endpoints, and credentials hardcoded instead of referenced from your environment — entirely in your browser. Nothing is sent anywhere. It's not a substitute for reading the server's source code, but it catches the most common red flags in seconds.
Frequently Asked Questions
No. It only performs static pattern-matching against the config text and any source/README you paste in — nothing is executed, and nothing leaves your browser.
The JSON block you'd put under mcpServers in claude_desktop_config.json, or a single server entry from any client (Claude Code, Cursor, Windsurf) that uses the same shape — command/args/env for local servers, or a url for remote ones.
These are heuristics, not a verdict. A filesystem server legitimately needs a path, and some scoped access is expected. Use the findings as questions to answer for yourself, not an automatic reject.
No. It can't see inside the actual server binary or detect logic bugs like a confused-deputy flaw. Pair it with reading the source, checking recent issues, and pinning versions.