AI For Asset Managers: UK Lessons From Bloomberg's New Enterprise MCP For AI Agents

What is MCP in AI?

MCP stands for Model Context Protocol. Anthropic introduced it on 25 November 2024 as “an open standard that enables developers to build secure, two-way connections between their data sources and AI-powered tools.”

In plain terms, MCP is a common plug between an AI agent and a source of data or a tool. Instead of building a custom connection for every system, a data provider publishes one MCP service and any compatible agent can ask it questions in a standard way.

What did Bloomberg launch on 29 September 2026?

Bloomberg announced Enterprise MCP from New York and London on 29 September 2026. It describes it as “an AI access layer for Data License Plus (DL+), Bloomberg's next-generation Data License offering.”

Through it, a firm's agents can work with licensed data across “more than 100 million securities and over 50,000 fields”, covering “pricing and reference data across asset classes, company fundamentals and structure, economics and alternative data”.

The first release includes ready-made skills. Bloomberg's examples are skills to “identify outliers in a set of tickers” and to “review trades that breach accepted price deviation thresholds”, along with sanctions exposure screening.

Two further steps are planned. Bloomberg intends to “extend Enterprise MCP with real-time data”, citing intraday monitoring, pre-trade checks, risk refreshes and exception handling. It is also building a solution for Terminal subscribers that would “provide the same governed and attributable access to Bloomberg data found in ASKB”, the Terminal's conversational interface.

Why does data access matter more than the model?

Bloomberg's own framing is direct. Tony McManus, Global Head of Enterprise Data and Indices, said “the bottleneck facing financial institutions has shifted from model capability to data readiness.”

Other providers are moving the same way. When Google launched its Gemini agent on 8 October 2026, it said Gemini for Financial Services draws on data from FactSet, LSEG and S&P Global as well as a firm's own repositories, and shows “full data lineage for easy auditing”.

For an asset manager, this means the useful question is no longer which model to pick. It is which data an agent may see, under whose licence, and how anyone can check later what it used.

Where does AI for asset managers in the UK start with market data?

Most of the early value for a boutique asset manager sits in operations, where the same checks are repeated every day:

• Price checks. An agent compares the prices used in valuations with an independent source and lists any outside tolerance, with the evidence for each.
• Position and cash reconciliations. An agent matches the firm's records with the administrator's and the custodian's, and explains each break. This is reconciliation automation applied to the fund.
• Trade reviews. An agent flags trades priced away from the market for a person to look at, much like Bloomberg's own example skill.
• Board and investor packs. An agent pulls figures, checks them against last month and drafts notes for the owner of the pack. See MI and board packs.

In each case the agent prepares and a person approves. Nothing is posted, sent or changed until someone with the authority to do so has looked at it.

What controls should sit around an agent's data access?

Bloomberg says it “validates a firm's entitlements before returning data on any agent request, under standard Data License rights”, and that clients control the agents and applications that connect, including models and prompts. That covers the data provider's side. The firm still owns everything after the data arrives.

1. Check the licence first. Confirm that your data licence covers use by AI agents, and for which purposes.
2. Start read-only. Let the agent read the data it needs and nothing else.
3. Log every request. Record what data each agent asked for, when and why, so any output can be traced back.
4. Approve every change. A person signs off before anything enters the books, a report or a client document.
5. Write it down. Keep a short record of what the agent does, who owns it and the manual fallback.

Connecting agents to outside tools also brings supply-chain risk. Our engineering team covers it in our Playbook guide to MCP tool security.

How can a boutique asset manager start without a large data team?

An agent does not need a live data feed on day one. Many useful checks run perfectly well from the exports a firm already produces each morning. BraivIQ, an AI agency in London for financial firms, starts with a 14-day Proof Run on up to three of those exports, read-only, and finds every mismatch with its likely cause.

If the numbers are there, one workflow goes live in 60 days, and connections to data services such as an MCP layer can follow once the controls are proven. For firms comparing workflow automation in London and across the UK, see the workflows our agents prepare. Our senior team learned these controls building surveillance for a global investment bank's trading floor.