Bloomberg launches Enterprise MCP so AI agents can use licensed market data
Bloomberg said Tuesday it launched Bloomberg Enterprise Model Context Protocol, an AI access layer for Data License Plus that lets clients' AI agents discover, understand, and retrieve licensed Bloomberg data through a standardized MCP interface.
Finance firms already have AI agents that can talk, but those agents still choke on bare market numbers. Bloomberg is putting licensed data behind MCP with meaning, entitlements, and audit so an agent can answer a question about a position without a person hunting for the right field.
On Tuesday, 29 September 2026, Bloomberg announced Bloomberg Enterprise Model Context Protocol, which it shortens to Enterprise MCP. It is an AI access layer for Data License Plus, which Bloomberg shortens to DL+ and calls its next-generation data license. An access layer, here, is the door a client’s own software uses to reach that licensed data. The release went out on PR Newswire, datelined New York and London. The page stamps it Sep 29, 2026, 03:30 ET, which is 3:30 a.m. Eastern. Those lines are Bloomberg’s.
What the product is for. Bloomberg says it lets a client’s AI agents discover, understand, and retrieve licensed Bloomberg data across more than 100 million securities and over 50,000 fields, through one standard MCP interface. A security, here, is a stock, bond, or other instrument a firm can hold. A field is one labeled item in that record, such as a price. MCP, the Model Context Protocol, is a shared plug so an agent can call a data service instead of a person hunting for the number. An agent, here, is software that takes steps, not only a chat reply. Bloomberg says the point is to move firms from AI experiments to workflows they can run every day. More than 100 million and over 50,000 are Bloomberg’s counts.
The gap Bloomberg says it is closing. Most tools can fetch a number. Fewer can tell an agent what that number means, how it was calculated, or whether it is the right one for the job. Enterprise MCP pairs semantic context and AI-ready metadata with workflow Skills. Metadata is the description attached to a field: what it means, how it was calculated, when it applies, and the qualifiers that make a number readable, such as the exchange, the currency, the kind of price, the period, and the as-of date. Bloomberg’s example is a last price that, on its own, does not say what currency it is in or what kind of price it is. Semantic search lets an agent find a field by describing it in plain language, instead of guessing Bloomberg’s short field codes. Entity resolution does the matching on the instrument side: a name, a ticker, or a description maps to the right security and to the company that issued it. Those lines are Bloomberg’s.
What sits behind that one interface, as Bloomberg lists it. Pricing and reference data across asset classes, company fundamentals and how a company is structured, and economics and alternative data, spanning DL+ Live and the Qube Datastore, which Bloomberg shortens to QDS. One tool discovers securities. One tool discovers fields. Bloomberg says that pair covers the content set, rather than a separate tool for each asset class. New content shows up as a new capability for agents that are already connected, not as a new integration to build. Those lines are Bloomberg’s.
Skills are reusable procedures. Bloomberg says clients will be able to adopt them and extend them. The procedures it names for that set cover pulling a universe of securities as of a point in time, adjusting prices for corporate actions, and reading revision history. A corporate action is an event such as a split or a dividend that changes a price series. The initial release, Bloomberg says, will include Skills that spot outliers in a set of tickers, review trades that break an accepted price-deviation threshold, and check securities and trades for possible sanctions exposure. More workflow Skills are to come. “Will include” is the release’s wording. It does not name a client already running those Skills.
Who is allowed to see what. Bloomberg checks a firm’s entitlements before it returns data on any agent request, under standard Data License rights. An entitlement is the slice of Bloomberg data that firm has licensed. Requests are answered from the Operational Datastore, which Bloomberg shortens to ODS. Tool-level limits cap how many securities and fields one call can return. Bloomberg hosts and manages Enterprise MCP. The client controls the agents that connect to it, including the models, the prompts, the instructions, and the other steering. A prompt is the instruction the firm gives its own model. Bloomberg says clients can license the content domains they need under those same rights, and that Data License clients can browse content at data.bloomberg.com. The release does not print a price.
Tony McManus, Global Head of Enterprise Data and Indices at Bloomberg, said the bottleneck has shifted from model capability to data readiness as generative AI use in financial services has grown. He said Bloomberg has seen agents that reason well and still cannot answer a basic question about a position, because the data reaching them does not say what it represents. He said connections were never the hard part. Readiness is, and that is what Enterprise MCP was built to solve. A position, in that sentence, is what the firm holds. That quotation is his, in the release.
Where Bloomberg says it sits in the protocol itself. It says its engineers are the main authors of proposals that add enterprise governance controls to MCP, and that let a server show different tool sets to different clients. Governance, here, means rules on who can call what. Bloomberg says it started and leads the MCP Financial Services Interest Group, and that it holds seats on the board and the Technical Committee of the Agentic AI Foundation, which keeps MCP as an open standard. Those lines are Bloomberg’s. The release does not name the other seats.
What Bloomberg says is next. It plans to extend Enterprise MCP with real-time data, for workflows that need a fresh price: watching the day as it happens, checks before a trade, refreshing risk, and handling exceptions. It is also delivering a forthcoming product for Terminal subscribers, so the same governed and traceable access that ASKB provides can sit in a person’s daily desktop work. ASKB is the Terminal’s conversational AI, as Bloomberg names it. Governed means the access still follows the firm’s license. Traceable means the answer can be tied back to the data it used. The release does not give a day for the real-time extension or for the Terminal product. Those lines are a plan.
In plain terms, Bloomberg opened a standard plug so a client’s own AI agents can ask for licensed market data and get the meaning with the number. The catalog it describes is more than 100 million securities and over 50,000 fields. The firm’s license still decides what comes back, and Bloomberg keeps a limit on each call. The client still picks the model and the instructions. The first Skills Bloomberg names are outlier checks, trade-price reviews, and sanctions checks. Real-time data and a Terminal version are plans, not a date.
The picture is a black and amber desk graphic of that stack. A client agent sits at the top. Enterprise MCP is the middle layer. Data License Plus is the base. The card names MCP, more than 100 million securities, more than 50,000 fields, and entitlements. It is a diagram of the launch. It is not a photograph of a Bloomberg terminal.