Toolport

Toolport is a free, open-source local MCP gateway that lets multiple AI agents share the same configured servers. It reduces token usage by up to 91% through on-demand tool searches. It is for developers using Claude, Cursor, VS Code, Codex, or ot...

Toolport

About Toolport

Toolport is a free, open source MCP (Model Context Protocol) gateway that sits between AI agents and MCP servers. Instead of dumping every tool definition into context on each request, it exposes a small set of meta-tools the agent searches on demand. The project is MIT licensed, runs fully local, and supports Windows, macOS, and Linux.

Review

Toolport addresses two problems that show up once you connect several MCP servers to an AI agent. The first is configuration sprawl: each client wants its own JSON file with API keys pasted in as plain text. The second is context bloat, since every server sends its full tool list on every request. The gateway approach handles both, and the security features are a side effect of putting a proxy in the path anyway.

Key Features

  • Meta-tool search: agents query a few small tool definitions instead of loading hundreds of schemas upfront, which cuts token overhead per request.
  • Local secret storage: API keys live in your OS keychain rather than in client config files.
  • Tool fingerprinting: the gateway flags rug-pulls and tool poisoning, and marks injection-looking tool output as data rather than instructions.
  • Approval gating: destructive calls pause until you approve them from the desktop app, with approvals bound to the exact arguments you reviewed.
  • Code mode: agents run a whole tool sequence as one sandboxed server-side script instead of round-tripping every call through the model.
  • Broad client compatibility: supports 33 clients including Claude, Cursor, VS Code, and Codex, plus a headless Docker gateway for server deployments.

Pricing and Value

Toolport is free and open source under the MIT license. There are no paid tiers or hosted services; everything runs locally and nothing routes through the developer's servers. The project had a separate Teams launch in July 2026 for sharing MCP servers team-wide, but its current pricing is not defined in the available material.

Pros

  • Measured token savings: benchmarks on a 415-tool catalog show definition overhead dropping 99.5%, and a 14-server setup costs ~886 tokens per request versus ~165k without the gateway.
  • One setup serves every client, so you configure each server once and all agents share it.
  • Security checks are automatic: tool fingerprinting, injection-output detection, and approval gates come with the gateway rather than requiring separate configuration.
  • Supports the latest MCP spec (2026-07-28) on both sides while older clients and servers keep working.

Cons

  • The search call itself adds latency to turns that use many tools; the maker acknowledges this tax scales with tool usage, though he estimates it stays cheaper than flat loading.
  • No reviews or third-party validation exist yet, so real-world reliability outside the maker's own benchmarks is unverified.
  • Not well suited for single-server setups where context bloat is minimal; the gateway adds an extra moving part that won't pay for itself at that scale.

Toolport fits developers who run multiple MCP servers across several agents and want to cut token costs without changing their workflow. It's also a reasonable choice if you want local secret storage and a checkpoint for destructive tool calls. Teams on a single client with one or two servers probably won't see enough overhead savings to justify the extra infrastructure.



Open 'Toolport' Website
Get Daily AI Tools Updates

Your membership also unlocks:

700+ AI Courses
700+ Certifications
Personalized AI Learning Plan
6500+ AI Tools (no Ads)
Daily AI News by job industry (no Ads)

Join thousands of clients on the #1 AI Learning Platform

Explore just a few of the organizations that trust Complete AI Training to future-proof their teams.