The MCP Multiplexer

One URL. All your MCP servers.

MCPlex lets you register your MCP servers, bundle them together, and expose a single endpoint that works with any MCP-compatible client. Stop editing config files every time you add a server — manage everything from one place.

The problem

Config file sprawl

Every MCP client needs its own config. Add a server? Edit claude_desktop_config.json, .cursor/mcp.json, and whatever else you use. Remove one? Do it all again.

No central management

Your MCP servers are scattered across configs with no single view of what's connected, what's working, or how to share a setup with your team.

API keys everywhere

MCP servers that need API keys force you to store secrets in plain-text config files on every machine. That's a credential management nightmare.

How MCPlex works

Register your MCP servers

Add any HTTP SSE MCP server. API keys are AES-GCM encrypted and stored securely — they never sit in a config file.

Bundle them together

Group servers into bundles. A "dev tools" bundle might combine GitHub, Linear, and your internal API. A "research" bundle might combine web search and knowledge base servers.

Connect with one URL

Each bundle gets a single MCP endpoint URL. Point your client at it — all tools from every server in the bundle are available immediately. Add or remove servers from the bundle anytime, no client-side changes needed.

Why it matters

Change servers, not configs.Add a new MCP server to a bundle and every connected client picks it up instantly. No restarts, no config edits.

One URL across all your tools.Claude Desktop, Claude Code, Cursor — configure once with the bundle URL. Switch tools without rebuilding your MCP setup.

Secure by default.OAuth 2.1 protects every endpoint. API keys are encrypted at rest. No secrets in local config files.

Share your setup.Bundles have public pages showing what's included. Useful for teams standardizing their MCP tooling or for sharing curated server collections.

Works with

Claude Desktop

Claude Code

Cursor

Any MCP client

Codemode

Let your AI do more in a single step

Normally, your AI assistant calls tools one at a time — ask a question, wait for a response, ask the next question, wait again. With Codemode enabled on a bundle, your AI can compose multi-step workflows that run all at once in a secure sandbox.

Without Codemode

Call tool AwaitRead resultwaitCall tool BwaitRead resultwaitCombine

Each tool call is a separate round trip. Multi-step tasks take multiple exchanges.

With Codemode

Describe the full taskrunGet the combined result

Your AI writes a short workflow using all available tools, executes it in one shot, and returns the final answer.

Faster answersTasks that would take five back-and-forth exchanges complete in one. Less waiting, more doing.

Sandboxed executionWorkflows run in an isolated environment. Your tools are accessible, but nothing else is — safe by design.

One toggle to enableFlip codemode on for any bundle. Your existing MCP endpoint stays the same — the codemode endpoint is an additional URL for clients that support it.