Quantradin
A simulated trading platform with an MCP server that allows users to connect AI agents for testing trading rules and running bots, without involving real funds.
Timeline (last 90 days)
Quantradin is a practice account where users or connected AI can describe rules and test them on historical U.S. stock and option prices.
Quantradin's key design features include no look-ahead functions, options filled at the bid-ask rather than the midpoint, and surviving strategies continuing to run as simulated bots.
On average, before deploying each bot, the Agent first runs about 17 backtests.
31 of the 40 bot pauses came from one account, and 29 of the 41 risk-control adjustments came from another account, indicating that these behaviors are highly concentrated among a few users rather than a general trend.
Read-only keys exist on the platform; holders can view, pause and close positions, but cannot open new positions or start bots.
On Reddit, a developer publicly shared the operating data of a simulated trading platform called Quantradin, which has an MCP server allowing users to connect AI Agents (mainly Claude) to test trading rules and run bots, in an environment that does not involve real funds.
The system only records calls that reach the server, not behavior in which the Agent reads a tool description and decides not to attempt it, which is a logging blind spot.
In their statistics, the developer excluded an anomalous account that performed 630 backtests with no other operations, because it would swamp the overall distribution.
Within five weeks, no account attempted to overstep its authority with a read-only key (opening new positions or starting bots), so zero rejections could mean the Agent understood the permission boundaries, or could mean users never requested higher permissions; from the audit table alone the two cannot be distinguished.
Within five weeks, 23 accounts generated a total of 421 calls, including 239 backtest submissions, 41 risk-control limit settings, 40 bot pauses, 71 Alpha backtests/signals/attachments, 9 deployment previews, 14 bot deployments, 5 bot retirements or mode switches, and 2 manual orders, with zero rejections.
All Facts (10)
Quantradin is a practice account where users or connected AI can describe rules and test them on historical U.S. stock and option prices.
50%UnverifiedQuantradin's key design features include no look-ahead functions, options filled at the bid-ask rather than the midpoint, and surviving strategies continuing to run as simulated bots.
50%UnverifiedOn average, before deploying each bot, the Agent first runs about 17 backtests.
50%Unverified31 of the 40 bot pauses came from one account, and 29 of the 41 risk-control adjustments came from another account, indicating that these behaviors are highly concentrated among a few users rather than a general trend.
50%UnverifiedRead-only keys exist on the platform; holders can view, pause and close positions, but cannot open new positions or start bots.
50%UnverifiedOn Reddit, a developer publicly shared the operating data of a simulated trading platform called Quantradin, which has an MCP server allowing users to connect AI Agents (mainly Claude) to test trading rules and run bots, in an environment that does not involve real funds.
50%UnverifiedThe system only records calls that reach the server, not behavior in which the Agent reads a tool description and decides not to attempt it, which is a logging blind spot.
50%UnverifiedIn their statistics, the developer excluded an anomalous account that performed 630 backtests with no other operations, because it would swamp the overall distribution.
50%UnverifiedWithin five weeks, no account attempted to overstep its authority with a read-only key (opening new positions or starting bots), so zero rejections could mean the Agent understood the permission boundaries, or could mean users never requested higher permissions; from the audit table alone the two cannot be distinguished.
50%UnverifiedWithin five weeks, 23 accounts generated a total of 421 calls, including 239 backtest submissions, 41 risk-control limit settings, 40 bot pauses, 71 Alpha backtests/signals/attachments, 9 deployment previews, 14 bot deployments, 5 bot retirements or mode switches, and 2 manual orders, with zero rejections.
50%