Twelve Data

The data layer for financial AI.

Real-time and historical market data for stocks, forex, crypto, ETFs, and fundamentals. 100k+ instruments, one API, one MCP server. Built for the people training models, running backtests, and shipping agents.

We ship the data. You bring the model. No proprietary models on this page yet, just the raw material and the tooling to plug it in.

→ Live demo + full app · API docs · Free API key

---

Featured dataset

twelvedata/financial-world-model

A reproducible financial ML dataset built on the Twelve Data API. ~50 large-cap stocks plus macro tickers (SPY, VIX, TLT, sector ETFs) across three timeframes: 1min, 1h, 1day.

Three output formats, one pipeline

• Parquet bars with causal technical indicators
• Instruction JSONL (prompt/label pairs) for LLM fine-tuning
• Trajectory JSONL / Parquet (rolling windows) for world-model and RL training

Built for research you can trust

• Idempotent daily updater with gap detection and backfill
• Leakage guards. Unit tests assert no future data in indicators
• Time-based train/val/test splits baked into dataset configs
• Split and dividend detection with automatic re-backfill
• Optional candlestick PNGs for multimodal training

from datasets import load_dataset

ds = load_dataset("twelvedata/financial-world-model")

For LLM fine-tuning on market data, time-series foundation models, and world-model / predictive-dynamics research.

---

MCP server

Market data as tools for Claude, Cursor, and any MCP client. The centerpiece is u-tool, an AI-powered universal router over 100+ Twelve Data endpoints. Describe what you want in English; u-tool uses vector search and GPT-4o to pick the endpoint, fill in parameters, and call it.

One tool, the entire Twelve Data ecosystem.

Connect it to Claude Desktop or any MCP client

{
  "mcpServers": {
    "twelvedata": {
      "command": "npx",
      "args": [
        "mcp-remote", "https://mcp.twelvedata.com/mcp",
        "--header", "Authorization:${AUTH_HEADER}",
        "--header", "X-OpenAPI-Key:${OPENAI_API_KEY}"
      ],
      "env": {
        "AUTH_HEADER": "apikey YOUR_TWELVE_DATA_API_KEY",
        "OPENAI_API_KEY": "YOUR_OPENAI_API_KEY"
      }
    }
  }
}

Things an agent can now answer

• Show me Apple stock performance this week
• Calculate RSI for Bitcoin with a 14-day period
• Get Tesla's financial ratios and balance sheet
• Find top-performing tech ETFs

Install options and source: github.com/twelvedata/mcp

---

API quickstart

If you're not using MCP, the REST API is one call away.

import requests

r = requests.get(
    "https://api.twelvedata.com/time_series",
    params={"symbol": "AAPL", "interval": "1day", "apikey": "YOUR_KEY"},
)
print(r.json())

Full reference: twelvedata.com/docs

---

Use cases

• Fine-tune LLMs on market data — Instruction JSONL is ready for SFT.
• Train time-series foundation models — Trajectory format, causal features, leakage-checked.
• Build financial agents — MCP u-tool gives one interface to the whole API.
• Backtest and research — Parquet bars with reproducible time-based splits.

---

Roadmap

More asset coverage and datasets are on the way. Reference notebooks and baseline models will follow. Contributions and feedback welcome — open an issue on the MCP repo or the dataset page.

---

twelvedata.com · API docs · Dataset · MCP server · GitHub · X