mcp-toolbox

This MCP server bundle provides several pre-built tools that Claude Code can call via standard input/output JSON-RPC. Each tool is built to run as a separate process and communicates over stdio, so every invocation starts a fresh process. You can access the tools inside the running Docker container with docker exec commands that target the pre-built tool entry points. For example, mcp-nixos is exposed via a Python module that starts the mcp_nixos.server, tailwind-svelte-assistant runs as a Node.js script at run.mjs, context7 is pulled in via npx, and agent-framework runs a Node.js server built under dist/mcp/server.js. Use Claude Code to register these tools by providing the appropriate command and arguments, so Claude can invoke them on demand for code search, documentation, or quality checks.

Prerequisites:

• Docker installed and running
• Access to the MCP Toolbox repository (this project)

Installation steps:

1. Install Docker on your system (follow platform-specific instructions from docker.com).
2. Clone the MCP Toolbox repository: git clone https://github.com/your-org/mcp-toolbox.git
3. Build and run the Docker container (examples assume a Makefile is provided as shown in the README): cd mcp-toolbox make build make run
4. Verify tools are available inside the container: docker exec -i mcp-toolbox ls -l /app/tools
5. (Optional) Persist tool data by enabling docker_volume: true in your deployment configuration, so tool data lives under hosts/servers as described in the README.

• Tools communicate via JSON-RPC over stdin/stdout. Each invocation should be treated as an isolated process.
• If you enable docker_volume: true for a tool, its data and native dependencies are stored on the host in the servers/ directory, enabling native execution outside the container.
• You can register tools with Claude Code using commands like claude mcp add <tool-name> -- docker exec -i mcp-toolbox <appropriate-entrypoint>, or by editing the Claude Code MCP settings JSON under ~/.claude/settings.json as demonstrated in the README.
• For context7 and the Node-based tools, ensure the required runtimes (Node, Python) are available in the container image or on the host if you choose native execution.