Share Model Context Protocol Servers Across 22 Clients with Toolport Gateway

Reducing Token Tax with Lazy Discovery

Every time an AI client connects to standard MCP servers, the entire list of available tools is injected into the LLM context. Toolport replaces this behavior with lazy discovery mode. Instead of exposing the full catalog, the gateway advertises four compact meta-tools to the agent: toolport_status, toolport_search_tools, toolport_call_tool, and toolport_fetch_result.

This approach reduces tool-definition overhead by 96% (measured on a frontier model), dropping the overhead for 62 tools from ~24,000 tokens to just 900. On a large catalog of 415 tools, the overhead reduction reaches 99.5%, resulting in up to 91% fewer total tokens consumed during task execution.

Single Config Across 22 AI Clients

Toolport acts as a centralized manager for all downstream MCP integrations. Instead of copying-and-pasting configuration snippets into Cursor, VS Code, Zed, Claude Code, and Codex separately, you configure each server once. Secrets are stored securely in your operating system's native keychain rather than in plain-text client configuration files.

Supply-Chain Security & Human-In-The-Loop

Because Toolport sits directly on the local execution path, it implements active defense mechanisms: tool integrity checks to detect schema changes ('rug-pulls'), content defense to flag prompt injections in tool returns, and human-in-the-loop approvals that pause destructive tool calls (e.g., database drops) until confirmed.