Modal MCP Toolbox
# Modal MCP Toolbox 🛠️
[](https://smithery.ai/server/@philipp-eisen/modal-mcp-toolbox)
A collection of Model Context Protocol (MCP) tools that run on Modal.
This let's you extend the capabilities of your LLM in tools such as [Goose](https://block.github.io/goose/) or the [Claude Desktop App](https://claude.ai/download).
<a href="https://glama.ai/mcp/servers/ai78w0p5mc"><img width="380" height="200" src="https://glama.ai/mcp/servers/ai78w0p5mc/badge" alt="Modal Toolbox MCP server" /></a>
## Tools
- `run_python_code_in_sandbox`: Let's you run python code in a sandboxed environment.
- `generate_flux_image`: Generate an image using the FLUX model.
## Demo
### Flux Image Generation

### Python Code Execution

## Prerequisites
- A [modal account](https://modal.com/signup) and a configured modal CLI.
- [UV](https://github.com/astral-sh/uv?tab=readme-ov-file#installation)
- A client that supports MCP. Such as the [Claude Desktop App](https://claude.ai/download) or [Goose](https://block.github.io/goose/)
This runs against your modal account, so you will need to have a modal account and be logged in.
## Installation
Installation depends on the client that uses the MCP. Here is instructions for Claude and Goose.
### Claude
Got to `Settings > Developer` in the Claude Desktop App. And click on Edit Config.

Add the config for the mcp server. My config looks like this:
```json
{
"mcpServers": {
"modal-toolbox": {
"command": "uvx",
"args": ["modal-mcp-toolbox"]
}
}
}
```
### Goose
Go to `Settings` and Click on Add.

Then add an extension like in the screenshot below.
The important part is to set command to:
```
uvx modal-mcp-toolbox
```
The rest you can fill in as you like.

### Installing via Smithery (not working currently)
To install Modal MCP Toolbox for Claude Desktop automatically via [Smithery](https://smithery.ai/server/@philipp-eisen/modal-mcp-toolbox):
```bash
npx -y @smithery/cli install @philipp-eisen/modal-mcp-toolbox --client claude
```
TDQS
Scored across 2 tools
The two tools have completely distinct purposes: one generates images using a specific AI model, while the other executes Python code in a sandboxed environment. There is no overlap in functionality, and an agent would never confuse these tools.
Both tools follow a consistent verb_noun pattern with snake_case naming: generate_flux_image and run_python_code_in_sandbox. The naming is clear, descriptive, and follows the same convention throughout.
With only 2 tools, this server feels extremely thin for a 'Toolbox' name that suggests broader utility. The tools are unrelated (image generation vs. code execution), making the server feel like two separate utilities bundled together rather than a coherent toolbox.
As a 'Toolbox,' there are significant gaps in coverage. The server lacks tools for common utility tasks like file operations, data processing, or other AI models. Even within the narrow domains represented, there are no complementary operations (e.g., no image manipulation tools to accompany generation, no code analysis tools to accompany execution).