Snowflake, the company of the AI Data Cloud (NYSE: SNOW), has introduced Cortex AI for Financial Services, a package of capabilities and partnerships designed for banks, asset managers, fintechs, and insurers to unify their data ecosystem and deploy models, apps, and AI agents on top of that data with security and regulatory compliance. The message comes with two clear points: bringing AI to where the data already lives — Snowflake — and making it interoperable with external agents and platforms via a Managed Model Context Protocol (MCP) Server, now in public preview.
“The financial sector has always been a pioneer, but it coexists with fragmented data, robust compliance, and the need for rigorous governance,” says Baris Gultekin, VP of AI at Snowflake. “By bringing AI where the data lives and enabling secure interoperability with remote agents, we facilitate critical use cases in highly regulated industries.”
The approach has both a technical layer and a business layer. On one side, Cortex AI adds components for data scientists, analysts, and business users (coding agents, AISQL functions, AI-driven extraction/transcription, and a conversational interface called Snowflake Intelligence). On the other side, a MCP Server connects internal and third-party data within Snowflake with agent platforms such as Anthropic (Claude), CrewAI, Cursor, Devin (Cognition), Agentforce (Salesforce), UiPath, or Windsurf, among others, to create “context-aware” experiences that act upon governed information.
What is Cortex AI for Financial Services (and why does it speak regulatory language)?
Cortex AI for Financial Services is born with an “enterprise-ready” and sector-specific focus. The suite aims to accelerate complex tasks —market analysis, quantitative research, fraud detection, customer service, claims management— and do so with security controls that fit into regulated frameworks. The idea is that an entity could, for example:
- Combine structured data in Snowflake (market history, tick data, portfolios, risk) with unstructured data (research, earnings call transcripts, document analytics) without leaving the governed perimeter.
- Invoke AI agents — internal or third-party — that read exactly what they’re permitted to (based on permissions), reason with that context, and execute actions within the workflow (create a note, mark a task, feed a workflow).
The data ecosystem accompanying this suite includes leading providers such as:
- Structured (via Sharing of Semantic Views, GA “coming soon”): CB Insights, Cotality™, Deutsche Börse, MSCI, Nasdaq eVestment®.
- Unstructured (via Cortex Knowledge Extensions, now GA): CB Insights, FactSet, Investopedia, The Associated Press, The Washington Post.
For Snowflake, the combination of industry-specific data (market analysis, expert research, news) with the client’s proprietary data in Snowflake enhances the accuracy and quality of AI outputs. In financial services — where a nuance of context can change a decision — context is everything.
The components: from coding agents to conversational AI
1) Data Science Agent (AI coding assistant)
Automates time-consuming tasks for data science teams: data cleaning, feature engineering, prototyping, and model validation. The goal is to speed up from raw data to production-ready models and ease bottlenecks in risk, forecasting, trading analytics, fraud, Customer 360, or underwriting.
2) Cortex AISQL (public preview)
Adds functions for extracting and transcribing with AI (also in public preview), enabling efficient processing of documents, audio, and images at scale. Think about claims analytics, customer support with attached documentation, investment digesting transcripts and reports… Much of the sector’s value resides in text and voice: AISQL puts these sources at the query’s fingertips.
3) Snowflake Intelligence (public preview)
For business, a conversational interface that lets you ask in natural language about data in Snowflake, and about third-party data/apps and connected agents. The promise is to democratize insights across the organization and reduce technical overhead hampering decision-making. Again, data stays in place; questions are asked to the data.
Managed MCP Server: the glue for an agent ecosystem (reducing ad-hoc integrations)
AI agents enhance LLMs with tools, workflows, and context. The common challenge: connecting them to enterprise systems often requires custom integrations that slow adoption. That’s where the Model Context Protocol (MCP) comes in, which has recently become a standard for LLMs to discover data, APIs, and services.
With the Snowflake MCP Server (public preview), a company can:
- Connect tools built on Snowflake (e.g., Cortex Analyst and Cortex Search) to external agents via a standard MCP interface, unifying the retrieval of structured and unstructured data.
- Expose proprietary data and third-party data shared on Snowflake Marketplace (via Cortex Knowledge Extensions) so remote agents can consume in real-time, without sacrificing security or governance.
Practically, the MCP Server allows agents from Anthropic, Augment Code, Amazon Bedrock AgentCore, Azure AI Foundry, CrewAI, Cursor, Devin (Cognition), Glean, Kumo, Mistral AI, Agentforce (Salesforce), UiPath, Windsurf, Workday, WRITER, and others to communicate with governed Snowflake data without each integration becoming a custom project. This is the difference between a “perpetual POC” and production.
Voices from the ecosystem: from Claude’s reasoning to multi-agent teams
- Anthropic (Jonathan Pelosi, Head of Industry, Financial Services): “With MCP, we can connect the governed data of each organization directly to Claude. Clients combine structured analytics and unstructured documents via Cortex Analyst and Cortex Search, maintaining enterprise security standards.”
- CrewAI (João Moura, co-founder and CEO): “The next wave involves orchestrating specialized agent teams. For them to work, secure and high-quality data is essential. Snowflake’s Managed MCP Server is the key pipeline enabling our crews to access, analyze, and act upon governed data. This turns multi-agent systems into an enterprise reality.”
- Cursor (Ricky Doar, Head of Field Engineering): “A coding assistant is limited by the context it has access to. A managed MCP Server like Snowflake’s creates a live data environment so tools like Cursor can consume context and write code that’s faster, more accurate, and safer for production.”
- FactSet (John Costigan, EVP, Chief Data Officer): “Making AI-ready data products available to clients in modern cloud environments is a decisive step toward unifying and enriching data.”
- Ramp (Ian Macomber, Head of Analytics): “With Cortex AI, we can analyze securely clients’ unstructured data; Ramp teams ask questions in plain English and get instant answers.”
- Salesforce Agentforce (Gary Lerhaupt, VP Product Architecture): “Extending interoperability of agents via MCP enables deeper connectivity and more intelligent agent experiences. Customers will discover and connect to Snowflake’s MCP server via AgentExchange.”
Sector-specific use cases (and why “live data” changes the game)
- Market analysis and quant research: researchers formulate queries (Snowflake Intelligence), request entity extraction from documents and transcripts (AISQL), prototype a model (Data Science Agent), and publish decision routes for agents that respond in real time via MCP.
- Fraud: unification of dispersed signals (structured/unstructured), automated features, and agents that alert or trigger workflows when patterns exceed a threshold.
- Customer service / claims: automatic extraction of data from forms, invoices, and audio; conversational AI that answers with evidence and cites documents, staying within governed data boundaries.
- Next-best action in banking and insurance: agents that consult updated context (saved by MCP) before recommending an action, reducing hallucinations.
Security and compliance: “AI goes to the data, not the data to AI”
All of the above rests on a premise that IT departments want to hear: privacy and permissions come first. With RTS (in the Slack environment) or MCP (in Snowflake and beyond), AI does not take the data elsewhere: it invokes the context at source, returns only what is needed, and respects permissions. For auditing and risk management, this architecture is simpler to defend than models based on “bulk export + parallel data lake”.
Availability (and fine print)
- Snowflake Cortex AI for Financial Services: available as a sector-specific suite with components in GA and public preview (see below).
- Snowflake MCP Server: public preview; designed for all industries seeking connected and interoperable AI.
- Cortex Knowledge Extensions: GA (for unstructured content).
- Sharing of Semantic Views: GA soon (for structured data).
- Cortex AISQL and extraction/transcription functions: public preview.
- Snowflake Intelligence: public preview.
As with all such announcements, Snowflake reminds that plans and timelines may change.
What does it mean for the industry (quick read)
- Fewer perpetual POCs: MCP standardizes the connection between agents and data; fewer custom integrations = more production.
- Governed data as an advantage: those who can align proprietary data, third-party data, and agents will reduce cycle times and errors caused by hallucinations.
- Conversational AI, but with roles: natural language fronts over traceable and explainable data; key for risk and compliance.
- Talent: Data engineers and scientists can move up the chain (less cleaning/repetition), while business teams access insights without tickets.
What to watch for in the coming months
- Parity between structured and unstructured data (performance, cost).
- Metrics of accuracy and latency in Snowflake Intelligence and AISQL at banking/insurer scale.
- Catalog of industry data on Marketplace and its alignment with Cortex Knowledge Extensions.
- Real adoption of MCP by large clients (public cases with audit and ROI).
- Security: guardrails for sensitive DMs/documents, incident playbooks, and native telemetry.
Conclusion: Contextualized AI at the core of data (with rules)
With Cortex AI for Financial Services and a managed MCP Server, Snowflake aligns three core elements: governed data, useful AI, and interoperability within an ecosystem of growing agents. For an industry driven by decisions based on context — and subject to regulators — the logic is clear: bringing AI to the data and not the other way around. If the market supports it, 2026 could be the year when agents with real data move from promise to standard.
Frequently Asked Questions
What is Snowflake Cortex AI for Financial Services, and how does it differ from “generic” AI?
It’s a sector-specific suite that combines AI capabilities (Data Science Agent, Cortex AISQL, Snowflake Intelligence) with data from leading financial providers (FactSet, MSCI, Nasdaq eVestment®, CB Insights, and others) and proprietary data in Snowflake, all while respecting governance and compliance.
What is the purpose of the managed Snowflake MCP Server?
It offers a standardized way for external agents (Anthropic, CrewAI, Cursor, Devin, Agentforce, UiPath, etc.) to discover and use governed data in Snowflake (own and third-party) without ad hoc integrations. It reduces integration time and risk.
What are the differences between Data Science Agent, Cortex AISQL, and Snowflake Intelligence?
- Data Science Agent automates data science tasks (cleaning, features, prototyping, validation).
- Cortex AISQL adds AI-powered extraction/transcription functions for documents, audio, and images (public preview).
- Snowflake Intelligence is the conversational interface enabling business users to ask questions in natural language about data in Snowflake and connected sources/apps (public preview).
Which external data providers are supported, and in what state?
For structured data (via Sharing of Semantic Views, GA coming soon): CB Insights, Cotality™, Deutsche Börse, MSCI, Nasdaq eVestment®. For unstructured data (via Cortex Knowledge Extensions, GA): CB Insights, FactSet, Investopedia, The Associated Press, The Washington Post.
Source: Snowflake