FactSet adds MCP for portfolio analytics
FactSet extends its MCP suite to cover portfolio analytics, giving buy-side investment teams natural-language access to performance, attribution, and risk data.
FactSet has added portfolio analytics to its Model Context Protocol suite, moving natural-language query access beyond the reference and earnings data already available via MCP. The limited release, announced 26 June 2026, targets buy-side investment professionals who currently need to open a thick-client application to pull yesterday’s attribution drill-down.
The architecture uses a semantic and metadata layer between the language model and FactSet’s analytics engine. When an agent asks a question in plain English, the layer translates it into a validated, pre-approved API call against FactSet’s pre-calculated analytics rather than letting the model touch raw data directly. Firms can restrict which portfolios, benchmarks, and time ranges a given agent is allowed to query — governance that previously required custom middleware.
The server runs inside each client’s private LLM environment, so portfolio data does not leave the firm’s infrastructure. The tooling includes developer support for custom agents and hooks into FactSet’s existing research solutions.
Portfolio Analytics extends a suite of five offerings on the FactSet Marketplace: an unstructured-data MCP, a vectorised data API, Event Hub, Intelligent Document Service, and now portfolio analytics. FactSet serves roughly 9,000 client organisations and 241,000 users.
The governance angle echoes a pattern across the MCP space: Mistral scoped its connector API keys and added tool-level toggles the same week, and Runlayer raised $30M specifically to build identity and policy enforcement for MCP agents. If you are wiring agents into sensitive data systems, giving each agent its own scoped key is the foundation all three of these products build on.