Model Context Protocol (MCP)
: An open protocol developed by Anthropic that standardizes how AI models connect to external data sources and tools. MCP provides a universal interface for AI agents to interact with databases, APIs, file systems, and other services through a client-server architecture. It replaces custom integration code with a standardized, discoverable tool interface.
Why It Matters for AI-Coded Apps
MCP servers extend AI agent capabilities significantly – and each extension is a potential attack surface. An insecure MCP server can expose sensitive data, allow unauthorized actions, or be exploited through prompt injection. As vibe-coded apps increasingly use MCP for AI integrations, understanding and securing MCP connections becomes critical.
Real-World Example
An MCP server provides database access to Claude Code. If the server does not implement access controls, the AI agent (and any prompt injection attack through it) can read, modify, or delete any data in the database. Proper MCP security requires authentication, authorization, input validation, and audit logging on the server side.
How to Detect and Prevent It
Authenticate all MCP connections. Implement authorization checks in MCP server handlers – do not trust that the AI agent will only make appropriate requests. Validate and sanitize all inputs from the AI client. Log all MCP tool calls for audit purposes. Run MCP servers with minimal database and filesystem permissions.
Frequently Asked Questions
Is MCP secure by default?
MCP provides the protocol structure but security depends on implementation. MCP servers must implement their own authentication, authorization, and input validation. The protocol itself does not enforce security controls – these are the responsibility of each MCP server developer.
What MCP servers are available?
Community MCP servers exist for databases (PostgreSQL, SQLite), file systems, GitHub, Slack, web browsing, and many other services. Official servers from Anthropic include filesystem and web search. Always audit third-party MCP servers for security before deploying them with your AI agent.
How is MCP different from function calling?
Function calling is a model-specific feature where the AI generates structured requests to predefined functions. MCP is a standardized protocol for discovering and calling tools across any AI system. MCP provides tool discovery, schema validation, and a universal interface; function calling is a lower-level model capability that MCP builds on.
Scan your app for security issues automatically
Vibe Eval checks for 200+ vulnerabilities in AI-generated code.
Try Vibe Eval