How to Vet an MCP Server: a Developer’s MCP Server Security Checklist

MCP servers expand AI assistants’ ability to access external tools and data, but documented incidents show they can also introduce risks through impersonated packages, vulnerable dependencies, excessive permissions, malicious updates, and prompt injection. A practical vetting process recommends preferring official vendor-hosted remote servers using OAuth, confirming that publishers match the relevant vendor, reviewing exposed tools and permissions for necessity, inspecting descriptions for hidden instructions, checking CVE history and maintenance activity, and carefully evaluating token-storage requirements; automated scanners such as mcp-vet can assist but should not replace manual review. After installation, users should apply least-privilege dedicated tokens, store secrets securely, pin versions and review updates, require human approval for destructive actions, treat all external content as potentially malicious, and monitor for expanded capabilities or OAuth scopes. Examples including a trojanized Oura clone, an OAuth remote-code-execution flaw, GitHub and Cursor prompt-injection demonstrations, a malicious Postmark package update, and an Amazon Q filesystem wipe illustrate that secure MCP use depends on both trustworthy software sources and cautious configuration, permissions, and interaction practices.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.