SmartStream, the data solutions provider for global financial institutions, has extended its use of agentic AI into collateral management with the launch of two Model Context Protocol (MCP) servers for its Smart Collateral platform, giving financial institutions a standards-based way to connect approved AI assistants directly to collateral data and workflows.
The move is designed to allow collateral teams to interrogate operational data using natural language and, where firms permit it, initiate actions through the same conversational interface.
The development comes as collateral management continues to become more operationally demanding. ISDA has highlighted the significant growth in collateral exchanged over the past decade as margin requirements have expanded, warning that manual processes, settlement delays and other operational friction can exacerbate liquidity pressures during periods of market stress. More recently, research by the association found operational issues to be a key obstacle to greater use of non-cash assets for variation margin.SmartStream’s approach aims to reduce some of that operational friction by changing the way users access information held within collateral systems. Instead of navigating different screens and reports to investigate agreements, eligibility, positions, margin calls or interest calculations, users can ask an AI assistant questions in plain language and receive answers based on Smart Collateral data.
Two MCP servers for Smart Collateral
The two servers provide different levels of access: one exposes Smart Collateral’s integration APIs as tools that an AI assistant can operate, while the other provides read-only access to the platform’s reporting database. The first MCP server makes Smart Collateral’s APIs available as a controlled toolset. It covers agreements, positions, margin activity and interest calculations and can pre-validate proposed collateral movements against existing eligibility and concentration rules before execution. Write actions are disabled by default and can be enabled individually, allowing firms to determine precisely which actions an AI assistant is authorised to perform.
“Collateral teams spend too much of their day hunting for answers and writing integration code,” says Jason Ang, Program Manager for Smart Collateral at SmartStream. “These MCP servers let firms point an AI assistant of their choosing straight at Smart Collateral, interrogating reporting data on demand, operating the integration APIs and, only where they choose to enable it, acting under their control.”
The second server provides read-only access to the Smart Collateral reporting database, allowing users to retrieve information conversationally or assemble dashboards without working through multiple reports. The infrastructure can also be used by development teams to generate integration code, reconciliation and exception-handling routines, and prototype automation workflows, with those outputs remaining subject to firms’ existing review, testing and deployment controls.The choice of the open MCP standard separates access to Smart Collateral from any single AI model or assistant, potentially providing an alternative to building bespoke integrations between individual AI models and operational applications. SmartStream says clients can connect MCP-compatible assistants that they have already approved rather than becoming tied to a proprietary AI ecosystem.
“It is an open, standards-based approach that keeps our clients in charge of their data, their workflows and their governance,” says Ang.
Extending SmartStream’s Agentic AI Strategy
The collateral initiative also builds on SmartStream’s broader move towards agentic automation within post-trade operations. In December 2025, the company previewed Smart Agents for Investigations, using domain-trained AI to triage, investigate and resolve reconciliation exceptions while documenting actions for audit and compliance purposes.
That was followed in June 2026 by the wider launch of Smart Agents, integrated with Smart Reconciliations and designed for both assistive and autonomous operation. The system allows institutions to determine how far automation should extend while retaining explainability and human oversight.
The new MCP servers apply a similar controlled-automation model to collateral, but with an architectural difference: rather than embedding the AI assistant itself within the application, SmartStream is opening a governed interface through which different compatible assistants can reach the underlying platform. That could prove particularly relevant as banks move from experimenting with generative AI towards connecting agents with live operational systems.
Subscribe to our newsletter
