mcp-toolbox-sdk-python

This MCP server wraps the Toolbox Python SDKs to expose Toolbox tools for GenAI applications. It enables you to fetch tool definitions from your running Toolbox instance, obtain Python objects or callables that represent those tools, and invoke them within your Python applications or orchestrators. The tooling supports integration with common frameworks like LangChain or LlamaIndex through the corresponding toolbox packages, while remaining usable in a framework-agnostic manner via toolbox-core. By using this MCP server, you can centralize tool definitions and authentication handling and then use the returned Python interfaces to execute API connectors, databases, or other Toolbox-managed tools from your code or automation flows.

To use it, install the toolbox-core package (or the broader toolbox suite) via the MCP server configuration, then connect your application to the Toolbox service URL. You can load tool definitions, instantiate tool objects or wrappers, and invoke them with the appropriate parameters. If you are building with LangChain, LangGraph, or LlamaIndex, prefer the corresponding toolbox-* package to ensure compatibility with those ecosystems. The setup typically involves creating a client instance, loading tools from Toolbox, and then calling the tools as standard Python functions or objects within your orchestration logic.

Prerequisites:

• Python 3.8+ installed on your system
• Network access to the Toolbox service endpoint
• A running MCP Toolbox service instance
• uvx utility installed (as used by this MCP server configuration) or a compatible Python runtime

Install and run the MCP server client:

1. Install uvx (or ensure it is available in your environment). If your environment uses a standard installation flow for uvx, follow those instructions to install the tool.

2. Install the toolbox-core package (the core, framework-agnostic SDK) via the MCP server run command described in mcp_config:

   • Ensure your Toolbox service URL and authentication token are configured (see env vars in mcp_config).

3. Start or run your MCP server configuration as defined in the mcp_config section of this repository. The MCP server will fetch definitions from Toolbox and expose usable Python tool interfaces to your application.

Example commands (illustrative; adapt to your environment):

# Ensure Python is installed
python3 --version

# Example: install uvx if applicable to your setup (follow your system docs)
# (This step varies by environment; provide your standard uvx installation command here)

# The MCP server configuration will be used as defined in the repository's mcp_config
# Typically you would ensure the environment variables are set, then run the MCP runner
export TOOLBOX_SERVICE_URL=https://your-toolbox-service-url
export TOOLBOX_AUTH_TOKEN=your-auth-token-or-placeholder

# Run the MCP server (via uvx as configured)
# The exact invocation depends on your environment; the config in this repo uses: uvx toolbox-core
uvx toolbox-core

4. Verify access by invoking a loaded Toolbox tool through Python, ensuring the tool definitions are loaded and the tool calls succeed.

Notes and tips:

• Ensure the Toolbox service URL (TOOLBOX_SERVICE_URL) and authentication token (TOOLBOX_AUTH_TOKEN) are correctly configured in the environment where the MCP server runs.
• If you are using LangChain, LangGraph, or LlamaIndex in your application, prefer the corresponding toolbox-<integration> package for best compatibility.
• The exact package versions should be aligned with your Toolbox service version to avoid compatibility issues.
• If you encounter connectivity or authentication issues, check network access, service URL correctness, and token validity. Review the Toolbox service logs for tool-loading errors.
• For local development, you can run a minimal Toolbox service instance and point the MCP server to it to test tool loading and invocation end-to-end.