MindStrat MCP
MindStrat's backtesting and live/paper trading surfaces cover exchange accounts such as Binance futures, with Binance's 0.05% per-side commission cited as the realistic cost setting for backtests before deploying a strategy to live or demo trading.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@MindStrat MCPBacktest a SuperTrend strategy on BTCUSDT 4h with realistic fees"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
MindStrat MCP
Drive the MindStrat desktop trading app from Claude Code — download market data, write and backtest strategies, run optimizations with robustness tests, and deploy to live trading, without leaving your terminal.
It ships with the app's manual, so the agent knows how each screen works, what every setting means, and how to write a strategy in MindStrat's MSF format that behaves in live exactly as it did in the backtest.
Install
You need three things:
The MindStrat desktop app, installed and running. This server talks to the app's local backend; it does nothing on its own.
Signed in to the app. The local engine is gated on your session.
uv, which runs the server.
Then, one command:
claude mcp add mindstrat --scope user -- uvx --from git+https://github.com/nachogarrid0/mindstrat-mcp mindstrat-mcpRestart Claude Code and check it connected with /mcp.
--scope usermakes it available in every directory. Without it Claude Code registers the server only for the folder you happened to be in.
Don't have uv?
# Windows (PowerShell)
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
# macOS / Linux
curl -LsSf https://astral.sh/uv/install.sh | shOpen a new terminal afterwards so it lands on your PATH.
Related MCP server: TradingView MCP Jackson
What you get
56 tools, grouped by the screen they drive: Market Assets, the Strategy Creator, the Optimizer, the Strategy Manager, portfolios and live trading. They call the same endpoints the app's own screens call, so the app owns every result — nothing here recomputes a metric.
The manual, which is the part that matters for writing strategies. The agent reads it on demand:
| What the app is and what each section does, the vocabulary of objects, and which state you share with the screen |
| Every control that changes a result: units, defaults, and which defaults lie |
| The rule the MSF format exists to protect, and the Live-Readiness Audit |
| The binding structure of a strategy and the full engine API |
| What the runtime rejects, and why |
| The annotated skeleton |
| Four complete worked strategies |
| The chart series schema |
| The tool catalogue, by screen |
Try it
With the app open, ask Claude Code:
What datasets do I have for BTCUSDT?
Write me a SuperTrend strategy on BTCUSDT 4h with an ADX filter, backtest it with realistic costs, and run the Live-Readiness Audit on it.
The second one is the real test. A strategy that passes that audit behaves in live the way it did in the backtest — which is the whole point of the format.
Two things worth knowing before you trade on it
Every cost default in the app is zero. Commission and slippage default to 0 in every surface, so an unconfigured backtest prices a frictionless market and every strategy looks better than it is. Set them before you believe a number — for Binance futures, 0.05% per side is realistic.
A strategy that exits on take-profit or stop-loss needs intrabar simulation. Without it, when one candle touches both levels the engine has to assume which came first, and it assumes the stop. Turning intrabar on can change a result substantially in either direction.
Both are explained in mindstrat://settings. The agent is told to read it.
Troubleshooting
BACKEND_UNREACHABLE — the app is not running. Start MindStrat and try again. Ask Claude Code to run backend_status to confirm.
401 / local.no_session — the app is running but you are signed out, or it restarted and lost the session. Sign in, or just click the MindStrat window to bring it into focus.
spawn uvx ENOENT — Claude Code cannot find uvx. Install uv (above) and open a new terminal. If it persists, use the full path: run where uvx (Windows) or which uvx and put that path in the command instead of uvx.
It connected but the tools do nothing — check you registered with --scope user. Without it the server only exists in one directory.
Live trading is off by default
Deploying, stopping and resizing a live strategy are disabled unless you opt in explicitly, by setting MINDSTRAT_MCP_ALLOW_LIVE=1 in the server's environment. Every one of them also requires a separate confirmation on the call itself.
environment="live" trades real money on your own exchange account. environment="demo" paper-trades against the testnet.
Stopping a live strategy tells it to stop deciding — it does not close an open position. Anything open stays open on the exchange, and its take-profit and stop-loss stop being watched. Close it yourself.
Privacy
The server runs on your machine and talks to your app over loopback. It holds no credentials of its own: it acts as whoever is signed in to MindStrat. Nothing is sent anywhere else.
This server cannot be deployed
Maintenance
Related MCP Connectors
MCP server exposing the Backtest360 engine API as tools for AI agents.
MCP server for OpenMM — exposes market data, account, trading, and strategy tools to AI agents
Mine, validate and paper-trade quantitative trading strategies. The engine runs on your machine.
Global stock research, ML forecasts, valuation signals, screeners & portfolio tracking in Claude
Related MCP Servers
- FlicenseNot gradedqualityBmaintenanceEnables Claude to control and read from the TradingView Desktop app, providing automated morning briefs, chart analysis, Pine Script development, and replay mode.99 npm-
- FlicenseNot gradedqualityBmaintenanceEnables Claude to control TradingView Desktop for automated morning briefs, chart analysis, Pine Script development, and session management via MCP tools.99 npm-
- AlicenseNot gradedqualityBmaintenanceLocal-first MCP server for building and running reproducible Backtrader strategies offline, with tools to inspect datasets, manage drafts, and execute backtests via controlled subprocess runs. It supports multiple hosts including Claude, Codex, OpenCode, and OpenClaw through stdio.MIT
- AlicenseNot gradedqualityBmaintenanceA Windows desktop control MCP server that gives Claude eyes and hands—screen capture with coordinate-grid overlays, pixel-perfect DPI-correct mouse/keyboard input via a Rust backend, and direct PowerShell command execution, enabling full local desktop automation through natural language.6 npm1MIT