T-Investments has implemented the Model Context Protocol (MCP) standard to connect third-party AI agents to its trading system. External algorithms now possess the technical capability to analyze market conditions, calculate tax liabilities, and place orders, including stop-losses. However, the architectural design excludes full autonomy: the agent executes any operation strictly following explicit user confirmation. This approach shifts the financial liability for algorithmic malfunctions or erroneous triggers directly onto the investor, minimizing the broker's legal risks when automating client strategies.
Key points
Integration of the MCP standard allows external AI algorithms to analyze markets, calculate taxes, and execute trades within T-Investments. The architecture requires explicit user confirmation for every operation, legally shielding the broker from algorithmic execution errors. Standardization eliminates the need to build custom APIs for individual developers, reducing infrastructure and inference support costs.
Strategic context
The bottom line
To stimulate development and attract talent, the company is launching the "Invest. Premarket" hackathon with a total prize pool of 2.5 million rubles. Participants will build tools for trading via the MCP protocol—ranging from specialized advisor agents and user interfaces to fully automated trading robots. Applications are open until November 11, with the in-person kickoff scheduled for November 21 at T-Bank IT hubs. From November 23 to December 3, teams will assemble MVPs online with mentor support. The finale takes place on December 5 in Moscow, where the grand prize totals 1 million rubles, alongside 500,000-ruble awards for winning solutions in categories covering investment AI agents, user experience, and trading robots.
What came before
Brokers previously had to sink capital into proprietary custom APIs and internal inference, but T-Investments shifted those compute expenses onto third-party builders by adopting the unified Model Context Protocol standard. The security perimeter remains tightly constrained: while external agents can calculate taxes or prepare stop-loss orders, requiring explicit client confirmation for every transaction strips the brokerage of legal liability for algorithmic slip-ups. Mandating human confirmation keeps autonomous software in the role of an advisory interface, leaving users with full financial liability for execution mistakes.