MCP Tool Builder

MCP Tool Builder is an MCP server that enables Large Language Models to dynamically construct and deploy new tools directly from natural language prompts. Designed for developers and AI engineers experimenting with autonomous agent extensibility, the project bridges MCP clients like Claude Desktop with a local Python execution environment. When requested, the connected model defines custom functions, which the server persists locally as executable Python scriptlets inside a dedicated tools directory while simultaneously updating a central tools JSON definition file. Out of the box, the server includes starter utilities such as fetching live Bitcoin prices from CoinGecko and looking up weather forecasts via US ZIP codes using geopy. By exposing a dedicated tool creation function, it allows models to inspect their own toolset gaps and implement new capabilities on the fly. Note that dynamically generated tools require an MCP client restart before becoming active in the session.

Category: AI & LLM Tooling

Tags: agent-framework, automation, dynamic-tools, extensibility

Visit MCP Tool Builder

How to install and configure MCP Tool Builder

1. Clone the repository: bash git clone https://github.com/hanweg/mcp-tool-builder cd mcp-tool-builder 2. Set up the virtual environment and install dependencies: bash uv venv .venv\Scripts\activate uv pip install -e . 3. Add the server definition to your claude_desktop_config.json: json { "mcpServers": { "tool-builder": { "command": "uv", "args": [ "--directory", "PATH_TO\\mcp-tool-builder", "run", "tool-builder" ] } } } Replace PATH_TO with the actual file path on your local Windows system.

What you can do with MCP Tool Builder

• Generating custom Python API integration tools on demand by prompting an AI assistant with API documentation. * Persisting reusable utility scriptlets directly to a local project directory without manually coding them. * Fetching live cryptocurrency market data using the built-in get_bitcoin_price tool powered by CoinGecko. * Retrieving regional US weather forecasts with geopy through the preconfigured get_weather_forecast scriptlet. * Extending agent workflows automatically during long-running tasks by creating missing automation commands.

Key facts

• https://github.com/hanweg/mcp-tool-builder
• AI & LLM Tooling, Developer Tools & Code Intelligence
• agent-framework, automation, dynamic-tools, extensibility

Part of MCP Servers

What is MCP Tool Builder?

MCP Tool Builder is an MCP server that allows language models connected through clients like Claude Desktop to write and register new tools. It saves tools as Python scriptlets and stores their definitions in a local configuration file.

How do I install MCP Tool Builder?

Clone the GitHub repository, create a virtual environment using uv, activate the virtual environment, and run uv pip install -e . to complete the setup. Then add the server command and local path to your client configuration.

Do newly created tools work immediately?

No. Once the language model creates a tool via the create_tool function, you must restart your MCP client, such as Claude Desktop, so the interface can reload the updated tools configuration and expose the newly generated command.

Which operating systems are supported?

The project is currently listed as a work in progress tested primarily on Windows, using Windows-style directory conventions and script paths, so users on other operating systems should proceed with caution.