Stochastly MCP

This small public launcher connects an MCP client to the server shipped with the Stochastly desktop app. The app must be installed, running, and licensed for Atelier or Portfolio. The package contains no Stochastly engine code.

Install with pip install stochastly-mcp or run with uvx stochastly-mcp. To try the unpublished source, install this directory with pip install ..

Configure a local stdio server with command stochastly-mcp and no arguments:

• Claude Desktop: add "stochastly": {"command": "stochastly-mcp"} under mcpServers in claude_desktop_config.json.

• Cursor: add the same entry under mcpServers in .cursor/mcp.json.

• VS Code: add "stochastly": {"type": "stdio", "command": "stochastly-mcp"} under servers in .vscode/mcp.json.

The server exposes 29 tools, among them node_catalog, compose_graph, backtest_graph, red_flag_check, adversarial_check, portfolio_combine and export_paper. The installed app runs them and checks the licence on every call: portfolio_combine needs Portfolio, every other tool Atelier or Portfolio.

Without the app, the launcher lists the same 29 tools (names, descriptions, input schemas and annotations from stochastly_mcp_data/tools_manifest.json, generated from the app's server), so a client can see what the server offers. Each description then starts with a note that the desktop app is required, and every call returns a structured answer with ok: false, the edition needed and a link to pricing. The manifest is metadata only.

For a nonstandard installation, set STOCHASTLY_MCP_SCRIPT to the installed app's stochastly_mcp.py. This is a local path, not a network endpoint.

Links

• MCP server overview: what the server does and which clients it supports.

• MCP server reference: the 29 tools, their inputs and which ones write.

• Documentation: the Stochastly desktop app, from data import to research papers.

• Comparison with other backtesting tools: how Stochastly differs from other backtesting software.

• Pricing: Atelier and Portfolio editions, 14-day trial.