Skip to main content

MCP Connector — Use Claude with Atomik

Atomik ships a Model Context Protocol (MCP) connector, so an AI assistant like Claude can work with your Atomik account directly from a chat: write a strategy, validate it against Atomik's strategy contract, backtest it on real futures data, and save it to your account — where it appears in your Strategy Builder, ready to activate.

If you already think through trading ideas with an AI assistant, this closes the loop: the same conversation that designs the strategy can now test it against two years of real data and hand it to you finished.

What Claude can do once connected​

ToolWhat it does
Get the strategy contractFetches Atomik's exact BaseStrategy interface so generated code is correct by construction
Validate a strategyChecks code in the same sandbox the live engine runs — free, no quota
Backtest a strategyRuns it against stored futures data (uses your plan's monthly backtest quota)
List available symbolsShows the instruments your plan can backtest
Save / list / get strategiesSaved strategies appear under My Strategies in the web Builder
Check usageYour remaining monthly quota

Two things it deliberately cannot do from chat: place trades, and activate a strategy for live trading. Going live always happens in the Atomik app, by you.

Connecting Claude​

  1. In Claude (claude.ai → Settings → Connectors), add a custom connector with the URL:
https://mcp.atomiktrading.io/mcp
  1. Claude sends you through Atomik's authorization page — log in to your Atomik account and approve. You're approving scoped access: strategy building and backtesting only, never trading.
  2. That's it. Ask Claude to build you a strategy.

You can see and revoke connected AI assistants any time in Settings → Connected Accounts → Connected AIs.

A session looks like this​

You: Build me an opening-range breakout for MNQ on the 5-minute — 9:30–9:35 range, one long and one short per day, 30-tick stop, 60-tick target, flat by 4 PM. Backtest it on the last 90 days.

Claude: fetches the contract, writes the Python, shows you the complete code, validates it, asks/confirms symbol-timeframe-range, runs the backtest, and reports the results — including the losing metrics.

You: Save it.

Claude: saves it and gives you the link — it's now in your Builder's My Strategies, ready to backtest further or activate on a paper account.

Quotas and requirements​

  • Requires an active Atomik subscription (any plan)
  • Backtests from chat consume the same monthly quota as the web Builder (Starter 30 / Pro 200 / Elite 500), with the same symbol and date-range limits by plan — see Strategy Builder
  • Validation is free and unmetered — Claude is instructed to validate before ever spending a backtest

Why this beats copy-pasting code​

Without the connector, an AI assistant guesses at Atomik's API from training data — plausible-looking code that fails on real interfaces. With it, the assistant fetches the actual current contract, validates against the actual sandbox, and tests on actual data. The difference is a strategy that works on the first save instead of the fifth debugging round-trip.

Next steps​