One Port for Every AI Tool — Stop Copying MCP Configs Into Five Apps

If you use AI tools, this mess is your life. MCP — the Model Context Protocol, the way AI tools plug into servers for files, web access, and more — is great, until you realize every client wants its own copy of the same server config. Claude Desktop, Cursor, VS Code, your CLI: five apps, the same setup pasted five times, and five places to fix when something changes.

There is a cleaner way: put the servers in one place and point everything at it. In the reel I ran two real MCP servers through a single gateway on one port. Here is the tool and the setup.

The tool: Toolport

Toolport (MIT, free and open source) is a local-first MCP gateway. You register each MCP server once in a registry.json, run one gateway process, and every AI client that supports MCP points at that one port (:8765) and shares them. It runs entirely on your machine — no login, no paid backend, no cloud account. (There is a paid “Toolport for Teams” tier for shared, governed setups; the local single-user gateway is 100% free.)

Set it up once

Install and start the gateway:

# install (see the repo for your platform)
# then start the gateway
toolport-gateway
# → gateway listening on http://localhost:8765

Register your MCP servers once in registry.json — for example a filesystem server and a fetch/web server:

{
  "servers": {
    "files": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/you/projects"] },
    "web":   { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-fetch"] }
  }
}

The gateway aggregates every registered server’s tools behind the single port. List them to confirm, and any MCP client can now reach all of them at once.

Point your AI clients at the one port

Instead of configuring files and web separately in every app, each client gets a single MCP entry — the gateway:

{ "mcpServers": { "toolport": { "url": "http://localhost:8765" } } }

Now Claude Code, Cursor, and any other MCP-aware client share the exact same tools. Add a server once, and every app has it. That is the whole point: set it up once, use it everywhere.

Honest notes

• “Every client” means every MCP-aware client. Toolport works with clients that support MCP. An app with no MCP support will not magically gain it.
• Some servers still need their own auth. A server that talks to a private API (Slack, GitHub) will still need its token — the gateway centralizes the wiring, not the credentials. It is not always zero-config.
• Config shapes vary by client. The exact mcpServers block above is illustrative; check your specific client’s MCP config format.

For anyone running more than one AI tool, this is the difference between maintaining one config and maintaining five. Set the servers up once, and stop pasting.