Snowflake Launches Cortex AI and MCP Server to Unify Financial Data for Agents

Snowflake: Can AI Agents Unlock Fragmented Financial Data?

Financial data and customer interactions sit in too many places. Snowflake is closing that gap with Cortex AI for Financial Services and a managed Model Context Protocol (MCP) Server, giving banks, insurers, and fintechs a direct route to context-rich AI without breaking security or compliance.

If you lead Finance or Customer Support, this means faster analysis, cleaner handoffs, and fewer custom integrations to maintain.

What Snowflake just launched

Cortex AI for Financial Services: AI models, apps, and agents that run where your data already lives in Snowflake, with enterprise controls.
Managed MCP Server (public preview): A standards-based way to connect Snowflake data and third-party sources (FactSet, MSCI, Nasdaq eVestment, The Associated Press) to external AI agents.
Snowflake Data Science Agent: Automates data cleaning, feature engineering, model prototyping, and validation for use cases like quantitative research, fraud detection, customer 360, and underwriting.
Cortex AISQL: New functions for extraction and transcription across documents, audio, and images for customer service, investment analytics, and claims management.
Snowflake Intelligence (public preview): A conversational interface that lets business users ask questions in plain language across structured tables and unstructured documents.
Data ecosystem: Structured and unstructured data from partners including CB Insights, Cotality, Deutsche Börse, MSCI, Nasdaq eVestment via Sharing of Semantic Views (GA soon) and Cortex Knowledge Extensions (GA).

Why this matters for Finance teams

Market analysis and quant research: Pull governed data and partner feeds into one place and let agents generate first-pass insights and backtests faster.
Fraud detection: Automate feature creation and model iteration with the Data Science Agent to shorten time-to-production.
Underwriting and customer 360: Blend internal risk signals with external sources; standardize scoring and explainability for audit.
Investment workflows: Use AISQL to extract entities from filings, transcripts, and news at scale, then route signals into analytics pipelines.

Why this matters for Customer Support and Claims

Unified case context: Search structured tickets and unstructured attachments, emails, and call transcripts from one interface.
Faster resolution: AISQL handles document parsing and audio transcription so agents spend time on decisions, not data prep.
Self-serve insights: With Snowflake Intelligence, non-technical teams can query policies, knowledge bases, and performance metrics in natural language.
Stronger controls: Keep data residency, PII handling, and logging centralized inside Snowflake while enabling external agent connections through MCP.

How the MCP Server changes integration work

Teams used to build one-off connectors for every agent and app. The managed MCP Server standardizes how large language models and agent platforms connect to Snowflake data, APIs, and services.

It links Cortex Analyst and Cortex Search with outside agents through a single MCP interface, unifying structured and unstructured retrieval. Supported platforms include Anthropic, Augment Code, Amazon Bedrock AgentCore, CrewAI, Cursor, Devin by Cognition, Glean, Kumo, Mistral, Salesforce's Agentforce, UiPath, Windsurf, Workday, and WRITER.

Who's in the data mix

Financial services firms can connect proprietary data with third-party sources from partners such as FactSet, MSCI, Nasdaq eVestment, CB Insights, Cotality, Deutsche Börse, The Associated Press, Investopedia, and The Washington Post. Sharing of Semantic Views is coming to GA, and Cortex Knowledge Extensions are GA for unstructured content.

Security, compliance, and control

Baris Gultekin, VP of AI at Snowflake, emphasized that bringing AI to where the data already lives-while enabling secure interoperability with remote agents-helps highly regulated institutions run business-critical use cases without adding risk.

Jonathan Pelosi, Head of Industry, Financial Services at Anthropic, added that MCP now connects governed data directly to Claude via Cortex Analyst and Cortex Search. Enterprises can apply advanced reasoning across both analytics and documents while meeting security standards.

Real-world traction

Ramp uses Snowflake Cortex AI to process large volumes of unstructured customer feedback. Teams can ask questions in plain English and get instant answers, accelerating decision-making across the company.

Fast path to value (30-day checklist)

Week 1: Identify 2-3 priority use cases (e.g., chargeback fraud, claims triage, earnings parsing). Map required data sources and access controls.
Week 2: Enable MCP Server, connect core Snowflake datasets and 1-2 Marketplace shares. Configure Cortex Knowledge Extensions for unstructured data.
Week 3: Prototype with Snowflake Intelligence and AISQL. Define prompts, guardrails, PII masking, and approval flows. Set evaluation metrics (accuracy, handle time, false positives).
Week 4: Pilot with 10-20 users. Log decisions and feedback, run shadow mode vs. baseline, and prep retraining loops. Establish monitoring and audits.

Key facts at a glance

Cortex AI for Financial Services helps deploy AI models, applications, and agents with security and compliance controls.
The managed MCP Server connects proprietary and third-party data (FactSet, MSCI, Nasdaq eVestment, The Associated Press) to agent platforms.
Integrations include Anthropic, CrewAI, Cursor, Salesforce's Agentforce, and Windsurf for context-rich agents and applications.
Snowflake Data Science Agent automates data prep and model lifecycle tasks for research, fraud, customer 360, and underwriting.
Cortex AISQL adds extraction and transcription for documents, audio, and images across support, analytics, and claims workflows.
Snowflake Intelligence (public preview) gives business users a conversational way to query structured and unstructured data.
Snowflake MCP Server links Cortex Analyst and Cortex Search to external agents via a standards-based interface.