One unified gateway for all your MCP servers.

Stop duplicating configs across IDEs and clients. Route local and remote MCP servers through one embedded Go control plane with Market, knowledge base, health checks, and live inspection.

Download v1.4.2 for All downloads

Example MCP clients

Transport and authentication support depend on the client version; verify an actual tool call.

Ollama

LM Studio

Cursor

VS Code

Claude

Codex

Features

Designed for production MCP workflows

MCPBox brings control, security, and intelligence to your AI workspace.

Project-centric orchestration

Group MCP servers by project and toggle them without manually editing JSON snippets for every client.

Aggregation endpoint

MCPBox handles initialization itself, aggregates tools, resources, and prompts, then routes calls back to the right backend.

Operator visibility

Health checks, audit logging, live inspection, and a clear project overview make it easier to understand why a client cannot see a server.

Integration flow

Use Market / Catalog to sync integrations from an external JSON manifest and attach them to projects quickly.

Skills and file processing

Connect Skill versions to projects, read instructions through MCP and run approved scripts in prepared Docker environments.

Knowledge base

Local Bleve full-text search or optional external RagBox ingestion and retrieval.

Quick Start

You can run MCPBox from source or just download the binary and start working right away.

Choose your path: build `MCPBox` from source for development or download a ready-to-run release.

Start `MCPBox`, create the administrator in the terminal and sign in at `http://127.0.0.1:38181/`.

Create a project and add local `stdio` or remote `HTTP streaming` servers.

Copy the project endpoint and connect Claude, Cursor, Codex, LM Studio, or a local Ollama workflow.

bash — MCPBox

# download release
MCPBox-macos-arm64
# allow run if needed
chmod +x MCPBox
# start MCPBox
./MCPBox
# open in browser
http://127.0.0.1:38181/

Documentation

Open link

Docs Home

Start with the public MCPBox documentation homepage.

Open →

Open link

Getting Started

Install, run, and connect MCPBox to AI clients.

Open →

Open link

Skills and scripts

Catalog, imports, bindings, Docker and file transfer.

Open →

Open link

ChatGPT and local clients

Direct local connections; HTTPS, Funnel and OAuth for cloud access.

Open →

Open link

Market and Knowledge Base

MCP integration catalog, local Bleve and optional RagBox.

Open →

Open link

Developer Guide

Architecture, transport model, aggregation behavior, and operator surface.

Open →

Open link

GitHub Repository

Source code, releases, and project history.

Open →

Open link

Release v1.4.2

Skills, refreshed workspace and upgrade notes.

Open →

Interface

What MCPBox looks like in practice.

Fresh captures of current MCPBox in a separate demo installation: a running MCP server, Skills catalog and settings. Click an image to open it at full size.

MCPBox main screen — Main dashboard
Projects, server state, endpoint visibility, and key actions in one place.

MCPBox Skills catalog filtered by author — Skills library
Search Skills, import packages and browse installed versions separately.

MCPBox settings with System and Security navigation — Settings and access
Storage and RagBox, users, authentication and active sessions.

Ready to use

If you need one MCP URL per project and a clear operator dashboard, MCPBox already solves that job.

The repository, documentation, and product overview now live on one page, so it is easier to present, onboard new users, and use as the main public entry point.

Open GitHub View releases