Moody's AI
Moody's, a global leader in financial intelligence, wanted to make their proprietary datasets, covering millions of companies worldwide, queryable through AI assistants. Working with MarsBased, they built an MCP server that connects Moody's financial data to Claude, enabling natural language access without exposing sensitive data to the model directly.
The challenge
Moody's holds one of the world's most comprehensive financial datasets: ratings, investment analysis, scorecards, and research covering millions of companies across every sector and geography.
The challenge was access. Enterprise teams and analysts wanted to query this intelligence through AI assistants, conversationally and in natural language, but the data is sensitive, proprietary, and subject to strict access controls. Sending it directly to a large language model was not an option.
Moody's needed a secure, auditable bridge between their internal data platform and the emerging generation of AI assistants, without compromising data governance or compliance.
Our approach
We built an MCP (Model Context Protocol) server in TypeScript that acts as a controlled gateway between Moody's financial datasets and Claude via Anthropic's API.
The server exposes Moody's data as structured tools and resources that Claude can call on demand (ratings, financial summaries, investment guidance) without the raw data ever reaching the model directly. Every query is mediated, scoped, and logged.
In parallel, we developed an AR prototype for Meta Quest demonstrating multi-agent conversational AI in augmented reality, presented at the Corporate Innovation Summit at Mobile World Congress.
Key design decisions:
Data isolation: Sensitive financial records never reach the LLM. The MCP server resolves queries, applies access rules, and returns only what the model needs to answer the question.
Access control: Role-based permissions determine which datasets each user or integration can reach, enforced at the server layer.
Audit logging: Every query is recorded with user context, timestamp, and scope, meeting the auditability requirements of a regulated financial institution.
Tech stack
MCP layer
TypeScript MCP server exposing Moody's data as Claude-callable tools.
AI integration
Anthropic Claude API via the Model Context Protocol.
Security
Role-based access control, end-to-end encryption, and full audit logging.
AR prototype
Multi-agent conversational AI in augmented reality on Meta Quest.
Results
We delivered a production MCP server enabling Claude users to query Moody's financial intelligence (ratings, investment analysis, company scorecards) through natural language, without any sensitive data reaching the model directly.
The integration positions Moody's as a launch partner in Anthropic's ecosystem, making their data platform accessible through the next generation of AI assistants.
Production MCP server in use by Moody's enterprise teams.
Natural language access to financial datasets with full data governance maintained.
AR prototype showcased at the Corporate Innovation Summit at Mobile World Congress.
Moody's positioned as a pioneer in AI-native financial intelligence tooling.
This project is part of a broader, ongoing partnership with Moody's across multiple AI initiatives, building on two years of collaboration and establishing MarsBased as one of their trusted technical partners.
Work with usLet's work
Let's work
together
We work with a select number of clients each year. Let's see if there's a fit.
Contact usComplex project, worldwide distributed stakeholders, thin definition. And with all that, MarsBased pulled it off. Great maintainable code, on time and within budget.
Director of AI and Quantum, Moody's