# Stochastly Desktop [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/MaheChadeffaud/stochastly-mcp  
**GitHub Stars:** 0  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/stochastly-desktop

## Description
Local MCP access to Stochastly desktop nodes, diagnostics and research tools.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "stochastly-desktop": {
    "command": "uvx",
    "args": ["stochastly-mcp"],
    "env": {
      "STOCHASTLY_MCP_SCRIPT": ""
    }
  }
}
```

**Requires environment variables:** `STOCHASTLY_MCP_SCRIPT` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What Stochastly Desktop MCP server does

The Stochastly Desktop MCP server gives an MCP-compatible client access to the MCP server included with the Stochastly desktop application. It is a launcher rather than the Stochastly engine itself: the package contains no engine implementation and delegates tool execution to the installed desktop app.

The exposed interface contains 29 tools. Named examples include `node_catalog` for inspecting available nodes, `compose_graph` for building graphs, `backtest_graph` for running backtests, `red_flag_check` and `adversarial_check` for analysis checks, `portfolio_combine` for portfolio operations, and `export_paper` for exporting research output. The complete tool reference, inputs, and write behavior are maintained by Stochastly.

Reach for the Stochastly Desktop MCP server when an agent needs to work with a local Stochastly installation through an MCP client. It is not a substitute for installing the desktop application, and it does not provide a remote network endpoint.

## How it works

The launcher starts as a local stdio process using the `stochastly-mcp` command with no arguments. The MCP client sends tool requests to that process, which connects to the server shipped with the desktop app. The app performs the requested operation and checks the applicable license on every call.

Atelier or Portfolio is required for all tools. `portfolio_combine` additionally requires the Portfolio edition. If the application is unavailable or the needed edition is not licensed, the launcher returns a structured response with `ok: false`, identifies the required edition, and includes a pricing link.

When the app is not present, the launcher can still advertise the same 29 tools. It reads names, descriptions, input schemas, and annotations from `stochastly_mcp_data/tools_manifest.json`. This manifest contains metadata only, so listing tools does not make them executable.

## Setup and configuration

Install the published package with `pip install stochastly-mcp`, or run it directly with `uvx stochastly-mcp`. A source checkout can be installed with `pip install .` when testing unpublished source.

Configure a local stdio server in the client using the command `stochastly-mcp` and no arguments. Claude Desktop and Cursor use an entry with that command under their `mcpServers` configuration. VS Code uses a server entry with `type` set to `stdio` and the same command.

For a nonstandard desktop installation, set `STOCHASTLY_MCP_SCRIPT` to the local path of the installed app's `stochastly_mcp.py`. This setting expects a filesystem path, not a URL or other network endpoint.

## Tools and capabilities

The Stochastly Desktop MCP server provides tools covering several stages of quantitative research in the desktop application:

- Inspect available nodes with `node_catalog`.
- Build and evaluate graphs with `compose_graph` and `backtest_graph`.
- Run `red_flag_check` and `adversarial_check` against research work.
- Combine portfolios with `portfolio_combine` when the Portfolio edition is available.
- Export a paper with `export_paper`.

These names are representative rather than a complete inventory; the launcher exposes 29 tools in total. Consult Stochastly's MCP reference for each tool's parameters and whether it writes data.

## Limitations and notes

The Stochastly Desktop MCP server depends on the desktop app being installed and running for real execution. Licensing is checked per request, so access can fail even after the client has successfully discovered the tools. The package itself does not include Stochastly engine code.

The fallback manifest is useful for client discovery and schema inspection, but it cannot perform research operations. The supported setup is local stdio, and `STOCHASTLY_MCP_SCRIPT` does not convert the launcher into a remote service.

_Full upstream README: https://allmcps.com/mcp/stochastly-desktop/readme_