MCP Toolbox

MCP Toolbox is an MCP server that provides a broad collection of utility tools to enhance LLM capabilities across local and cloud environments. It connects LLMs to command-line execution, local file systems, web search engines, Figma designs, audio transcription, persistent semantic memory, and image generation APIs. Developers, designers, and general AI assistant users employ MCP Toolbox to expand their agents beyond basic text generation. The server enables models to inspect Figma components, execute shell commands, read and modify local project files, search the web via Tavily or DuckDuckGo, convert documents and web pages to Markdown with MarkItDown, process audio lengths and segment transcriptions, and store durable cross-session thoughts and memories. By packaging these capabilities into standard Model Context Protocol interfaces, MCP Toolbox equips MCP-compatible desktop and terminal assistants with comprehensive multi-modal system interaction tools.

Category: AI & LLM Tooling

Tags: framework, integration, toolkit

Visit MCP Toolbox

How to install and configure MCP Toolbox

1. Ensure you have uv installed, or use pip in Python 3.12+. 2. Open your client configuration file, such as Claude Desktop (claude_desktop_config.json). 3. Add zerolab-toolbox under mcpServers using uvx: json { "mcpServers": { "zerolab-toolbox": { "command": "uvx", "args": ["--prerelease=allow", "--python=3.12", "mcp-toolbox[all]@latest", "stdio"], "env": { "FIGMA_API_KEY": "your-figma-api-key", "TAVILY_API_KEY": "your-tavily-api-key", "DUCKDUCKGO_API_KEY": "your-duckduckgo-api-key", "BFL_API_KEY": "your-bfl-api-key" } } } } 4. Alternatively, install via Smithery by running npx -y @smithery/cli install @ai-zerolab/mcp-toolbox --client claude. 5. Restart your MCP client to load the tools.

What you can do with MCP Toolbox

• Interrogate Figma files, inspect UI component sets, retrieve styles, and read or publish team comments directly inside your assistant. - Execute local command-line instructions, search directories, and perform regex replacements across local code files. - Search web content using Tavily or DuckDuckGo APIs and save rendered HTML or converted Markdown to local disk. - Store, retrieve, and query cross-session persistent memories via local semantic search storage on macOS and Linux systems. - Download podcast audio from XiaoyuZhouFM, convert formats, check durations, and transcribe specific segment timestamp ranges. - Generate image assets using the Flux API (BFL) and directly save the resulting graphics files locally.

Key facts

• https://github.com/ai-zerolab/mcp-toolbox
• AI & LLM Tooling
• framework, integration, toolkit

Part of MCP Servers

How do I install MCP Toolbox?

You can install and run MCP Toolbox using uvx with the command uvx mcp-toolbox[all]@latest stdio. Alternatively, install it globally via pip with pip install mcp-toolbox[all] or run automated installation for Claude Desktop using Smithery with npx -y @smithery/cli install @ai-zerolab/mcp-toolbox --client claude.

What can MCP Toolbox do?

MCP Toolbox provides a modular set of tools for command-line execution, local file operations, Figma project interaction, MarkItDown document conversion, web scraping, Tavily search, Flux image generation, podcast audio downloading, audio transcription, and long-term semantic memory storage.

Which MCP clients work with MCP Toolbox?

MCP Toolbox works with any client supporting the Model Context Protocol over stdio or SSE transport. Tested clients include Claude Desktop, Cline, and Cursor. You can configure it by adding the uvx or python execution command to your client settings file.

Where does MCP Toolbox store persistent memories?

On macOS systems, memories are saved to ~/Documents/zerolab/mcp-toolbox/memory, which automatically syncs across Apple devices when iCloud Drive is enabled. On other operating systems such as Linux and Windows, memories are saved locally in the ~/.zerolab/mcp-toolbox/memory directory.