smart-mcp-proxy/mcpproxy-go

Supercharge AI Agents, Safely

What it solves

AI agents often face limitations when interacting with many tools, such as hitting API token limits or exceeding function caps in IDEs like Cursor. Additionally, connecting multiple Model Context Protocol (MCP) servers can introduce security risks like tool poisoning attacks.

How it works

MCPProxy acts as a single, secure endpoint that sits in front of your various MCP servers. It aggregates tools into a single retrieve_tools function, which significantly reduces token usage and improves response accuracy. It includes a security quarantine system that blocks new servers until they are approved, and supports pluggable security scanners (like Snyk or Trivy) to audit servers before they are trusted.

Who it’s for

Developers using AI-powered IDEs (like Cursor or VS Code) or desktop agents (like Claude Desktop) who need to scale the number of tools available to their AI models without hitting context window or API limits.

Highlights

  • Token Efficiency: Reduces token usage by up to 99% by aggregating hundreds of tool schemas into a single function.
  • Security Quarantine: Automatically blocks potentially malicious servers and allows for manual approval.
  • Pluggable Scanners: Supports Docker-based security scanners to normalize risk scores for new servers.
  • Cross-Platform: Single binary for macOS, Windows, and Linux with an embedded web UI.