# MCPBox > Control center for managing MCP servers with project-based organization, web UI, and audit logging. - **Type:** MCP server - **Install:** `agentstack add mcp-artursaleev-mcpbox` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [ArturSaleev](https://agentstack.voostack.com/s/artursaleev) - **Installs:** 0 - **Category:** [AI & ML](https://agentstack.voostack.com/c/ai-and-ml) - **Latest version:** 0.1.0 - **License:** Apache-2.0 - **Upstream author:** [ArturSaleev](https://github.com/ArturSaleev) - **Source:** https://github.com/ArturSaleev/MCPBox ## Install ```sh agentstack add mcp-artursaleev-mcpbox ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # MCPBox MCPBox is a Go-based control plane for MCP servers. It lets you group local and remote MCP backends by project, inspect what they expose, monitor traffic, and publish a single project URL for AI clients such as Claude, Cursor, Codex, LM Studio, and local Ollama-based workflows. Click any screenshot to open it in full size. ## Why MCPBox Managing several MCP servers quickly turns into a mix of JSON snippets, terminal tabs, and half-documented local setup. MCPBox gives you one embedded UI and one local binary for the operational side of that work. Key capabilities: - Project-based organization for local `stdio` and remote `HTTP streaming` MCP servers - One project MCP endpoint at `/mcp/{project_token}` that aggregates all enabled servers in the project - Built-in Knowledge Base / RAG collections with local indexing for code, text, CSV, XLSX, DOCX, PPTX, and text-based PDF - Live inspection of `tools`, `resources`, `prompts`, and nearby README files for local `stdio` servers - Health checks on create, update, manual check, and local server start - Audit log for MCP traffic and control actions - Project pause/resume and per-server enable/disable controls - Embedded Market / Catalog flow for syncing, installing, verifying, and uninstalling integrations - One-click Ollama and llama.cpp launchers for local MCP testing - Single-binary deployment with embedded React UI and local SQLite storage ## Quick Start 1. Download the binary for your OS from the latest release, or build from source. 2. Run `MCPBox`. 3. Open `http://127.0.0.1:38180/` if the browser does not open automatically. 4. Create a project. 5. Add local `stdio` servers or remote `HTTP streaming` servers. 6. Copy the project endpoint from the UI and connect your AI client to `/mcp/{project_token}`. For local Ollama testing: 1. Make sure `ollama` is installed. 2. Open a project that has at least one enabled MCP server. 3. Choose a local Ollama model in the UI. 4. Click `Launch Ollama`. For LM Studio: 1. Open a project that has at least one enabled MCP server or connected knowledge base. 2. Click `Launch Project`. 3. Choose `Add to LM Studio`. 4. Confirm the deeplink in LM Studio when your OS opens it. For local llama.cpp Web UI testing: 1. Make sure `llama-server` is installed and available in `PATH`. 2. Set `MCPBOX_LLAMACPP_MODEL` or choose a local `.gguf` model in the launch dialog. 3. Open a project with an enabled MCP server or connected knowledge base. 4. Click `Launch Project` and choose the llama.cpp action. 5. MCPBox starts `llama-server`, applies the project prompt as a system prompt, and opens the Web UI. ## Build From Source Requirements: - Go `1.26+` - Node.js and npm Build steps: ```bash npm --prefix html install npm --prefix html run build go build -o MCPBox . ``` Run from source: ```bash go run . ``` Default port: `38180` ## Documentation - [README-ru.md](./README-ru.md) - Russian user guide - [DEVELOPER.md](./DEVELOPER.md) - developer guide, API notes, and architecture - [DEVELOPER-ru.md](./DEVELOPER-ru.md) - Russian developer guide ## Notes - MCPBox uses a local SQLite database by default. - The current Knowledge Base / RAG implementation uses local full-text indexing with Bleve. It does not build or require embedding indexes at this stage. - The Docker runtime support in `1.2.1` is intentionally an MVP. Advanced container features such as compose-style orchestration, custom networks, and volume presets are not fully implemented yet. - The `go_install` package flow in `1.2.2` assumes the target integration can be installed through `go install module/path@version` and exposed through a concrete binary entry point. - Secret masking is applied in the UI and catalog install flow. Existing manually configured servers that still pass secrets directly in command arguments should be migrated to environment variables for full process-list safety. - The Ollama launcher is shown only when `ollama` is installed on the machine. - The llama.cpp launcher requires `llama-server`; project prompts are passed through `--system-prompt-file` and are also exposed through MCP `prompts/list` as `project_prompt`. - Local server inspection is available for `stdio` servers only. ## Source & license This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [ArturSaleev](https://github.com/ArturSaleev) - **Source:** [ArturSaleev/MCPBox](https://github.com/ArturSaleev/MCPBox) - **License:** Apache-2.0 Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/mcp-artursaleev-mcpbox - Seller: https://agentstack.voostack.com/s/artursaleev - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.