Modal MCP Toolbox

A collection of tools for your LLMs that run on Modal

python

23

GitHub Stars

python

Language

6 months ago

First Indexed

2 months ago

Catalog Refreshed

Documentation & install

Readme and setup notes from the catalogue, plus a client-ready config you can copy for your MCP host.

Installation

Add the following to your MCP client configuration file.

Configuration

View docs

{
  "mcpServers": {
    "philipp-eisen-modal-mcp-toolbox": {
      "command": "uvx",
      "args": [
        "modal-mcp-toolbox"
      ]
    }
  }
}

Modal MCP Toolbox lets you extend your LLM workflows by running MCP-enabled tools directly on Modal. It provides tools like running Python code in a sandbox and generating Flux-based images, accessible from MCP clients such as Claude Desktop App or Goose. This enables you to perform dynamic actions and computations within your conversations and improve your automation with MCP-enabled capabilities.

How to use

You use an MCP client (Claude Desktop App or Goose) to connect to the Modal MCP Toolbox. Your client will load the toolbox as an MCP server and expose the available tools inside your MCP session. Once configured, you can invoke the toolbox tools from your chat or prompts just like other MCP endpoints.

Two primary runtime configurations are shown for the toolbox. The Claude/Modal setup uses a local stdio server that you start via the uvx tool, and Goose uses the same approach. Every connection point is described below so you can reproduce the setup in your client.

Available tools include running Python code in a sandbox and generating Flux images. You can trigger these by selecting the corresponding tool in your MCP client and providing your inputs. This makes it easy to test code safely and produce visual outputs directly from prompts.

How to install

Prerequisites you need before installing the toolbox:

A Modal account and a configured Modal CLI (UV)
An MCP-compatible client such as Claude Desktop App or Goose

Install steps for Claude and Goose are shown in the configuration examples below. Follow the exact commands to ensure the toolbox is discovered as an MCP server by your client.

Note: An automated installation path via Smithery is not working at the moment, so use the manual configuration paths described here.

If you prefer to copy the exact config for Claude, use the JSON snippet shown in the Claude setup. It configures the MCP server named modal-toolbox to run with uvx and the modal-mcp-toolbox command.

Additional notes

Two explicit MCP configurations are provided for the toolbox, one for Claude Desktop App and one for Goose. Both use the same stdio-based runtime and command to start the toolbox: uvx modal-mcp-toolbox. Ensure you are logged into your Modal account when starting these configurations.

The toolbox exposes a minimal set of tools described as part of its capabilities: running Python code in a sandbox and generating Flux images. You can extend usage by combining these tools with other MCP prompts in your workflow.

Usage examples and tips

In Claude Desktop App or Goose, add a new MCP server using the provided stdio configuration and point it to the modal-mcp-toolbox. - Start the toolbox and then invoke the tools by name in your MCP prompts. - For Python sandboxing, supply the code you want executed within the input field and receive the sandboxed output. - For image generation, supply the Flux-based prompt or image instruction to generate visuals.

Available tools

run_python_code_in_sandbox

Run Python code in a sandboxed environment to safely execute user-provided scripts without affecting the host system.

generate_flux_image

Generate an image using the FLUX model based on input prompts or parameters.