Commentary by: Chris Edmonds | President, Fixed Income & Data Services, ICE

Any AI model is only as good as the data and methodology behind it. In fixed income, where evaluations span multiple segments and pricing is proprietary and complex, the quality of underlying data is a clear differentiator.

As financial firms apply large language models, users increasingly want to make decisions by asking a question in plain text and receive a multi-sourced answer within the AI assistant they already use. Commercial constraints have quickly become clear. A single question can burn enormous token volumes as models reason over raw, unstructured data, and paying for every unit of compute consumed rarely reflects the value an institution realizes. Cost is only part of the issue. For institutions, the challenge is to pair the flexibility of consumption-based pricing with predictability and value.

In this month’s release, the ICE Data Services MCP connector brings ICE’s proprietary fixed income evaluations inside AI assistants, starting with Claude. When a user asks a question in their LLM, the MCP connector maps each security to its proper evaluation frame. It then applies ICE's domain expertise and methodology to surface the relevant authoritative data, drawing on proprietary content that is structured so that an LLM can turn it into a fully sourced response.

In this way, the MCP connector can be thought of as a methodology-grounded knowledge layer built on the expertise of ICE’s evaluators - insight that historically, was mainly accessible through direct client engagement.

Unlike workflows that depend on shared infrastructure, evaluating a bond doesn't require your counterparty to be on the same platform. This is where ICE's position is distinct because we author and own the evaluation methodology, from the models to the framework that governs how each asset class is assessed.

For firms that already license ICE evaluated pricing, our MCP connector extends the value of that entitlement. In seconds, a portfolio manager can understand the drivers behind a day-over-day move on any position - rate, issuer, or bond-specific - before responding to a client inquiry. A middle-office valuation-oversight team can substantiate a price challenge with methodology-grounded data and context, and fund accounting can investigate a NAV outlier without leaving their workflow.

More broadly, the MCP connector is part of ICE’s Aurora strategy which aims to enhance the value of our data and solutions for clients through AI. Here, we offer a variety of engagement channels. Our native solutions include ICE Aurora for Desktop, an intelligent agent that adapts to your role, entitlements, and market access with agentic workflows. This rollout begins with ICE Chat on July 30 and over time, will apply AI capabilities across a range of tools that clients are already using. Our application of AI can also be seen in tools like ICE Compass - a platform that fuses a firm’s trading data with our pricing streams to sharpen pre-trade counterparty selection.

ICE’s MCP connector is built on the principle that users should pay for the value of data delivered rather than the volume of compute consumed. Instead of giving an LLM thousands of rows to interpret, it applies ICE's methodology to allow it to craft an answer and maximize efficiency. A request for evaluated prices across ~3,300 investment-grade technology bonds for example, returns a single value-weighted result, quoted before the query runs and aggregated across every user at the firm. In this way, ICE helps prevent useless token-maxing or runaway bills.

Data is valued differently, where ICE’s proprietary evaluated pricing (for example) is weighed more heavily than public reference data. In addition to the vanilla securities that ICE evaluates daily, our infrastructure, pricing models, and dealer connectivity mean we specialize in pricing more complex, thinly traded instruments across the fixed income universe.

For years, ICE has delivered its fixed income data through bulk files, APIs, streaming feeds and cloud-native delivery, and these remain the foundation of how institutions power their books, risk systems and pricing engines. The MCP Connector sits alongside these channels as a distinct option so that ICE's data and methodology is brought into a user's AI environment.

The MCP connector’s Phase 2 release on July 24 expands its foundation beyond FINRA TRACE and MSRB content by adding ICE’s U.S. Treasury and AAA municipal benchmark curves, end-of-day evaluations, enhanced transparency and price-verification context. In coming months we'll broaden content available in the knowledge layer, so the responses market participants receive from their AI assistants can be grounded in the same rigor, sourcing, and auditability that clients have always demanded from ICE.