QuantCraft Tools
QuantCraft Tools is a built-in set of read-only market tools the AI can call for you — in the QuantCraft AI chat, in an Agent AI Prompt node, and in AI Test / Optimization agents. Instead of guessing at prices or copying numbers into your prompt, the model fetches what it needs and reasons over the real data.
It is on by default. Manage it from AI → MCPs in the top menu: QuantCraft Tools (built-in) is the first row, with an Enabled for chat checkbox, alongside any custom MCP servers you add. That checkbox affects the chat only — AI Prompt nodes always have the tools.
What the tools do
| Tool | What it returns |
|---|---|
| get_ohlcv | Recent OHLCV bars for a symbol (up to 60). |
| compute_indicator | An indicator's latest value(s) — rsi, macd, bollinger, atr, adx, vwap, supertrend, and the rest of the chart set. Output lines match the chart (e.g. MACD → macd / signal / histogram). |
| get_fundamentals | Company financials — balance sheet, income statement, cash flow, earnings, ratios, per-share values, valuation and scores (Piotroski / Altman / Beneish) — as of a point in time. The model can ask for just the fields it needs. See Fundamentals. |
| run_screener | Runs a fundamentals screen and returns the matching symbols. See Screener. |
| get_account | Balance, equity, realized / unrealized P&L. |
| get_positions | Open positions. |
All six are read-only — the AI can look, never trade. Placing orders stays in the Agent Execution node.
In the assistant chat
Ask a data question and the model calls a tool mid-answer — you'll see a status such as Calling MCP (QuantCraft Tools): get_ohlcv while it runs. Examples:
- "What's AAPL's last close and its 14-day RSI?"
- "Screen for profitable large caps trading under 15× earnings."
- "How much buying power do I have, and what am I holding?"
get_ohlcv, compute_indicator, get_account and get_positions use your connected Trade account for live data. With no account connected they report that the data isn't available (get_account says "not connected") rather than failing. get_fundamentals uses your signed-in QuantCraft session (the same access that powers backtest fundamentals); signed out, it reports "not available".
In an Agent AI Prompt node
When an AI Prompt node runs, the model can call the same tools — scoped to that run:
get_ohlcv/compute_indicatorsee the run's own bars for the run symbol.get_fundamentalsis filing-date-correct for the current bar.get_account/get_positionsread the run's paper or live account.
So a prompt like:
"Use compute_indicator and get_fundamentals to decide whether this is a value entry. Answer true or false."
…lets the node gather evidence before it answers, instead of you pre-loading everything into the prompt text.
Notes:
- Works with QuantCraft Premium (Claude grant only), a connected Anthropic or OpenCode key, and self-hosted Custom LLM endpoints (OpenAI-compatible, with tool-calling). DeepSeek — own key or hosted — isn't supported for AI Prompt nodes.
- A test evaluates the node at every scheduled evaluation (the Scheduler's bar interval), and each tool round is a model call, so the AI call budget (500 calls per test) covers tool use too. Narrow the date range or use a coarser schedule if you hit it. There is no setting to change the budget or to switch tools off for a node.
run_screenerinside a run is capped at about 45 seconds per call (or stops when you press Stop); a partial result is returned and marked as such.- Each tool call shows in the node's Output tab with a one-line summary.
In Test / Optimization agents
AI Test and Optimization agents can use the same tools when Allow market data tools is ticked in the agent's settings (off by default). See AI Test and Optimization agents.
See also
- QuantCraft AI — connecting a model, AI Tools and MCP servers
- Agents — AI Prompt nodes, AI Builder, Test / Optimization agents
- Screener — the screens behind
run_screener