QuantConnect

What it is

QuantConnect's hosted platform: a browser IDE, cloud backtest and research nodes, a data library mounted beside them and a deploy button that runs the same algorithm live through a real broker. Algorithms are Python or C#.

Underneath is LEAN, the open-source engine QuantConnect also publishes on its own. The engine costs nothing; what the cloud sells is compute, cleaned data and the live-trading plumbing. This card is that cloud — you come here so as not to run any of it yourself.

Pricing

Paid plans are bundles of nodes sold per seat, and the pricing page opens on annual billing. Month-to-month: Researcher $84, Team $168, Trading Firm $480, Institution $1,272. Nodes are also sold separately — a B2-8 backtest node $14/month, an R1-4 research node $12, live nodes from $24 to $1,000 — so a pack price is a floor, not a ceiling.

Those four figures are the recommended packs rather than the entry price. A Researcher seat on its own is $10/month; the $84 pack is that seat plus one research node (1 CPU, 4 GB), one backtest node (2 CPU, 8 GB), one live node (1 CPU, 0.5 GB), one AI agent node and Bronze support. Seats carry the features, nodes carry the compute, and the two are billed separately, so the plan is assembled rather than chosen.

Data & coverage

US equities, options, index options and futures from AlgoSeek, plus forex, CFDs, crypto and international futures, tick to daily, with corporate fundamentals in every tier. Universe selection is written in code, not clicked in a screener. AlgoSeek's US equities start in January 1998 and are survivorship-bias-free, about 27,500 securities from the SIP feed.

Fundamentals are Morningstar's US dataset — about 900 fields from January 1998, excluding ETFs, ADRs and OTC stocks — and in autumn 2026 its whole history was replaced. Morningstar retired the feeds behind it, and QuantConnect rebuilt the history back to 1998 from the new ones. About a third of the values both generations carry differ, usually by more than 20%; coverage grows from 8,717 to 9,705 companies; 535 properties are new and 1,251 are gone, a retired one throwing when read. The old data dated ratio, growth, per-share and average values at the period end, while the filing behind them came a median of 63 days later — so old backtests could read figures before they were public, which live algorithms never could. The new data takes the filing date.

Downloads for a local LEAN are billed separately, and that is where budgets break. On a Researcher seat the US Equity Security Master is $600/year (it rises to $1,800/year at Institution), AlgoSeek's US equity minute history $980/month and its tick history $1,400/month, current coarse-universe files $240/year, Benzinga news $120/month, Quiver's congress and insider feeds $5/month each.

Integrations

Python or C#. An official MCP server — now shipped inside the VS Code Local Platform extension, with the standalone repo marked deprecated — lets an LLM create projects, run backtests and deploy live algorithms through the API. Live algorithms notify over Telegram, email, webhook and SMS, the last billed in QCC tokens at 1 USD per 100 (1 QCC domestic, 10 international per message).

Limitations

• The free tier is a demo of the cloud, not a free platform: no live trading — the live-node limit is literally zero — no second or tick data, no API, no CLI. What it does allow is one backtest node, one research node, 200 projects, 500 MB of workspace, 32 KB per file, 25 data-file downloads per backtest and 10,000 orders and insights in a single backtest, against 10 million on a Researcher seat. A universe strategy reaches that last one quickly.
• The data stays in the cloud. Moving the same algorithm to a self-hosted LEAN means buying the downloads above or sourcing all of it yourself.
• Fundamental backtests from before the Morningstar rebuild will not reproduce. The migration guide keeps the old data as the default on LEAN's master until 10 October 2026, then on a separate branch, and retires it for backtesting and live trading on 31 October 2026; the dataset page dates the switch from 23 September 2026 instead. The old data cannot be kept alongside the new, so re-run anything that ranks on fundamentals before trusting it.
• No screener, charting or portfolio UI. Everything is code.
• The abstractions — algorithm framework, universe selection, consolidators — are a real learning curve, and a strategy written against them moves to a self-hosted LEAN and to no other engine.

Alternatives

LEAN is the same engine on your own server, with your own data and broker login. Backtrader and Zipline-reloaded for a plain Python library, NautilusTrader for a faster event-driven core, QuantRocket for a self-hosted platform wired to Interactive Brokers.