Summary
LSEG’s MCP connector gives AI agents access to trusted, licensed LSEG content and services through a standardised interface designed for agentic workflows.
In agentic workflows, cost is driven not only by the user’s prompt, but also by the tools the model considers, the calls it makes and the data returned after each request.
LSEG’s MCP connector is designed to help reduce avoidable token use through clear tool instructions, precise search and resolution, and compact responses.
That can mean fewer misdirected calls, smaller model-facing payloads, faster workflows and better economics as usage scales.
Customers remain in control of the model, prompts, tools exposed and context passed forward. LSEG’s role is to make the data connection structured, governed and efficient.
AI agents are changing how firms use market data. As financial institutions move from AI experimentation to production, the challenge is changing. Firms do not just need models that can reason. They need models that can access trusted, permissioned and high-quality data efficiently, repeatedly and at scale. Rather than asking a person to find, export and prepare information, an agent can connect directly to trusted data sources, retrieve what it needs and use that information inside a workflow.
That is the promise of LSEG’s MCP connector: giving agents direct access to the breadth of LSEG content — including pricing, fundamentals, reference data, estimates, fixed income analytics, news and more — in a way designed for real-world, scaled use.
For market professionals, this matters because the most useful agentic workflows are rarely single-step lookups:
A portfolio manager may use an agent to screen issuers, compare valuation metrics, monitor news and review portfolio-level exposures.
A research analyst may move from a company to its instruments, estimates, fundamentals, pricing and peer comparisons.
A risk manager may need to understand exposures across portfolios, asset classes and scenarios.
A data or technology team may need to expose governed data to AI systems without adding unnecessary complexity or cost.
The value appears when an agent can move across data types and analytical steps without losing precision. But every step raises an important question: how much does the model have to read, decide and process before it produces a useful answer?
That is where token efficiency matters.
Token efficiency is not only about cost. It is also a usability issue. A workflow that requires fewer unnecessary calls and less irrelevant context can be faster, easier to scale and more consistent in production. Tokens are the units large language models process. In a tool-based workflow, the model may need to process not only the user’s request, but also conversation context, tool descriptions, schemas, tool outputs and data carried forward from previous steps. In a shallow workflow, that overhead may look modest. In a deeper workflow, where an agent searches, resolves, retrieves, compares and refines its answer over several turns, it can compound quickly.
LSEG’s MCP connector is engineered to help improve token efficiency by reducing avoidable token use, without reducing the breadth or depth of the data available. The aim is simple: help the model reach the right data faster, and make sure it only has to read what matters.
Why token efficiency matters
When an AI agent uses data, it usually follows a sequence: understand the question, decide what information is needed, choose a capability, retrieve the result and reason over what comes back. In many workflows, that sequence repeats.
Some token usage is valuable. It supports richer context, deeper analysis and a better answer. But some token usage is avoidable. If the model chooses the wrong tool, receives a result that does not answer the question or has to read large amounts of irrelevant structure around the data, the workflow spends tokens without improving the outcome.
That is why token efficiency is central to making AI agents practical at scale. A data connection is only valuable if agents can use it repeatedly across real workflows without the overhead of each interaction becoming a constraint.
Two design choices that make the difference
LSEG’s approach focuses on two places where unnecessary token usage often arises.
First, the model has to understand which data capability to use. In MCP, a tool is a capability the model can call — for example, to search for an instrument, retrieve a price, resolve a company identifier, return a dataset or access an analytical model. The clearer the tool instructions, the easier it is for the model to choose correctly.
Second, the model has to read the response. Financial data services can return large technical structures containing repeated field names, metadata, formatting and other information that may be useful for systems, but not always useful for model reasoning. If all of that is passed into the model, the model reads more than it needs.
LSEG’s MCP connector is built around reducing waste in both places: clearer tool instructions before the call, and more compact responses after the call.
Difference 1: clear tool instructions reduce unnecessary calls
The first point of difference is how LSEG describes and organises the tools available through its MCP connector.
A model connected to a large data environment may have many possible actions available. If those actions are described loosely or overlap in unclear ways, the model has to work harder to decide what to do. It may call a tool that is close, but not quite right. It may receive data that does not answer the question. Then it has to recover and try again.
Each wrong turn consumes tokens, adds latency and increases the chance of an inconsistent workflow.
LSEG reduces that uncertainty by making tool instructions clear, specific and aligned to the way financial market data is actually used. Each capability is described so the model can understand what it does, when to use it and what information it needs. Where precision matters — for example, identifying the right company, issuer, instrument, venue, benchmark or curve — search and resolution capabilities help the agent orient itself before committing to a deeper retrieval.
The cheapest unnecessary call is the one that never happens. If the model chooses the right path earlier, it avoids failed attempts, repeated requests and extra context that the model then has to read and correct for.
The practical benefit is fewer avoidable calls, faster workflows and improved token efficiency, while still giving agents access to the breadth of LSEG content.
Difference 2: compact responses reduce unnecessary reading
The second point of difference is what comes back after a tool is called.
A raw data response can be much larger than the information the model needs. For example, if a model asks for values across several companies or instruments, an inefficient response may repeat the same field labels, metadata and structural formatting for every record. The underlying data may be valuable, but the wrapping creates extra tokens.
LSEG’s MCP connector is designed to shape responses around what the query needs: the relevant records, the relevant fields and the right level of detail. The response is organised for model consumption rather than simply passing through a full raw object that the model has to read past.
This does not reduce the quality or depth of the data. It reduces avoidable overhead around the data.
The model reads less helping improve token efficiency.
In financial markets, efficient access cannot come at the expense of governance. LSEG’s MCP connector is designed to support that requirement by providing a structured connection to trusted, licensed content, while customers remain in control of the models, prompts, tools and context they choose to use.
How the workflow changes
Generic or less optimised data connection
- Model reviews unclear or overlapping tools
- Model may choose a tool that is close but not right
- Wrong or incomplete result comes back
- Model tries again, using more tokens
- Large raw object is passed to the model
- More avoidable token use, more latency, more friction
LSEG MCP connector approach
- Model reviews clear, specific tool instructions
- Model identifies the right capability earlier
- Relevant result comes back
- Response is shaped for the query
- Model receives the answer with less unnecessary wrapping
- Improved tokens efficiency, faster workflows, better economics
The important point is not that every token can or should be eliminated. The goal is to make sure the tokens that remain are useful. A justified token spend should support better reasoning, broader context or a stronger answer. It should not be absorbed by avoidable retries, ambiguous tool choices or repeated structural padding.
LSEG’s AI strategy is built around making trusted data and analytics available where clients need them, across platforms, models and workflows.
Why this matters for scale
In small experiments, inefficiency can be easy to miss. A model that makes one unnecessary call, or reads one oversized response, may still produce a useful answer.
But enterprise workflows are different. Agents may run across many users, portfolios, asset classes, data types and analytical steps. At that point, small inefficiencies compound. A little extra context in one turn becomes significant when repeated across thousands of interactions.
This is why the boundary between the model and the data matters.
LSEG does not run the customer’s model inference or control token usage. What LSEG can control is the quality of the tools exposed through its MCP connector, the structure of the data returned and the way LSEG content is made usable by AI systems.
Customers remain in control of how the workflow is assembled: which model is used, which tools are exposed, what prompts are written and what context is passed forward. LSEG’s role is to make the connection to licensed financial data as structured, governed and efficient as possible.
Breadth without the noise
The value of LSEG’s MCP connector starts with breadth: trusted market data, reference data, fundamentals, estimates, analytics, news and more, available directly inside AI workflows and subject to customer entitlements.
But breadth alone is not enough. For agentic AI to work in financial markets, that breadth needs to be precise, permissioned and efficient to use. If an agent has to hunt through unclear tools, reconcile ambiguous entities or process oversized responses, that breadth becomes harder to use efficiently.
LSEG’s MCP connector is built to make that breadth usable. Clear tool design helps the model choose the right path. Search and resolution help it identify the right entity, instrument or issuer. Compact responses help it focus on the answer rather than the wrapping. Together, these choices help improve token efficiency by reducing avoidable token use in the workflows where agentic AI creates the most value.
Token use should support better reasoning and richer answers. LSEG’s MCP connector is designed to make sure the rest — the overhead of reaching that data — is as small as it can be. Wherever clients choose to build AI workflows, LSEG’s MCP connector supports them through connections that are open, governed, scalable and designed for real-world use.
The breadth is the reason to connect. The engineering behind the boundary is what makes that breadth practical to use, repeatedly and at scale.
Legal Disclaimer
Republication or redistribution of LSE Group content is prohibited without our prior written consent.
The content of this publication is for informational purposes only and has no legal effect, does not form part of any contract, does not, and does not seek to constitute advice of any nature and no reliance should be placed upon statements contained herein. Whilst reasonable efforts have been taken to ensure that the contents of this publication are accurate and reliable, LSE Group does not guarantee that this document is free from errors or omissions; therefore, you may not rely upon the content of this document under any circumstances and you should seek your own independent legal, investment, tax and other advice. Neither We nor our affiliates shall be liable for any errors, inaccuracies or delays in the publication or any other content, or for any actions taken by you in reliance thereon.
Copyright © 2026 London Stock Exchange Group. All rights reserved.
