Who this is for: heads of AI and data at banks, asset managers and corporates deciding which licensed data their agents should reach first. Skip it if you need free market data; every provider here is a paid licence.
In short: In about twelve months every major financial data provider adopted the same agent connection protocol, MCP. Your agents can now call licensed data from inside Claude, ChatGPT, Copilot and the cloud marketplaces. They differ on how much they disclose about real use, how they price it, and how tightly they enforce entitlements.
Key points
- FactSet said in July 2026 that more than 10% of organic growth came from AI products, with 450+ MCP clients.
- S&P Global reported 500+ clients on its agent-ready API and MCP service in July 2026, up more than 70% in a quarter.
- LSEG had more than 200 customers engaged on its MCP offering by July 2026, less than a year after launch.
- Bloomberg's Enterprise MCP, launched 29 September 2026, checks entitlements on every request but discloses no adoption or pricing.
- Ask every provider about licence checks and logs, model-training restrictions, and pricing at ten times today's call volume.
How we ranked
| Criterion | Weight | Why it matters |
|---|---|---|
| Disclosed adoption | 35% | Client counts and usage growth, stated publicly, are the best available sign the integration works. |
| Breadth of AI surfaces | 25% | Whether your agents can reach the data from the tools your firm already uses. |
| Entitlements and audit | 20% | Every call should be checked against your licence and logged. |
| Pricing clarity | 20% | Consumption pricing is still being invented; clearer is better for budgeting. |
The list
1. FactSet
- Best for: buy-side and banking teams that want the provider with the most evidence behind its AI products.
- Why it ranks first: the only peer that quantifies AI's share of growth. In July 2026 it said more than 10% of its organic growth came from AI products that were close to zero a year earlier, more than 450 clients were on its MCP service under contract or trial, and call volume rose 13-fold in a quarter. It is available on all the major model platforms.
- Watch out for: its adjusted operating margin fell about three points year on year, which management put down to pay, technology and marketing spend. Expect that to show up in pricing.
- Pricing: enterprise agreements with minimum commitments plus consumption tiers.
- What would change its rank: a peer disclosing comparable revenue attribution.
2. S&P Global (Kensho)
- Best for: teams that use Capital IQ and want it inside their own agents.
- Why it ranks here: more than 500 clients on its agent-ready API and MCP service in July 2026, up more than 70% in a quarter, with API volume five times the prior quarter.
- Watch out for: the same quarter brought an earnings-per-share miss, despite the AI growth.
- Pricing: a mix of consumption and data-set pricing.
- What would change its rank: revenue attribution like FactSet's.
3. LSEG
- Best for: teams already on Workspace who want data through Claude or ChatGPT.
- Why it ranks here: more than 200 customers were engaged on its MCP offering by July 2026, less than a year after launch, with tool calls up nearly five times between May and June. Its chief executive said MCP comes up in about a third of its AI-related commercial discussions.
- Watch out for: its finance chief said MCP would not materially affect 2026 results. Early days, by its own account.
- Pricing: usage-based for MCP and API; premium add-ons for research agents.
- What would change its rank: published client numbers for its research agent.
4. Moody's
- Best for: credit, lending and compliance teams that want workflows, not just data.
- Why it ranks here: it ships whole agent workflows rather than raw data: credit memo generation, peer comparison, sanctions and ownership screening, running inside Claude and listed on AWS Marketplace, over more than 600 million entities. By July 2026 it reported more than 100 MCP and API connections in use or trial.
- Watch out for: it has not disclosed AI revenue or client counts. Its strongest figure is a cohort one: customers buying a standalone AI product retain at 97% and grow about twice as fast, which partly reflects which customers buy first.
- Pricing: use-case deals today, with consumption trials and a move toward enterprise licences announced.
- What would change its rank: a published commercial model and client count.
5. Bloomberg
- Best for: large institutions that want agents to reach Bloomberg data inside their own environment.
- Why it ranks here: the most ambitious architecture. Its Enterprise MCP, launched 29 September 2026, lets clients' own agents discover and retrieve licensed data across more than 100 million securities and 50,000 fields, with entitlement checks on every request and audit trails.
- Watch out for: no adoption or pricing disclosed, no accuracy benchmarks published for its Terminal assistant, and it self-hosts rather than listing in the major model catalogues.
- What would change its rank: any disclosed usage.
Also callable today
| Provider | What agents can reach | Notable term |
|---|---|---|
| Dun & Bradstreet | Commercial graph, with the D-U-N-S number as the identity anchor | Available in ChatGPT and Codex since June 2026 |
| MSCI | Index and ESG data via a Claude connector | "Hundreds of clients" reported in April 2026 |
| Verisk | Insurance data via Claude | MCP included in the base subscription; client pays the tokens |
| Nasdaq | Data plus Verafin's financial-crime agents | Separate upcharge for the MCP layer |
| ICE | Fixed-income evaluations | Claude only; priced on data delivered, not compute |
Left out, and why
- Fitch. No AI platform of its own disclosed; it supplies content into a third-party platform instead.
- Morningstar. An assistant inside its own advisor workflow, not yet a service other agents can call.
Three questions for any provider
- Is every agent call checked against our licence, and can we see the log?
- Does the licence forbid using the data to train models? Get it in writing.
- How will pricing work at ten times today's call volume? Agents use far more data than people.
Common questions
Which financial data providers support MCP for AI agents?
FactSet, S&P Global (Kensho), LSEG, Moody's and Bloomberg lead the list, and Dun & Bradstreet, MSCI, Verisk, Nasdaq and ICE are also callable by agents today.
Which financial data provider has the strongest AI adoption?
FactSet ranks first as the only provider that quantifies AI's share of growth: more than 10% of organic growth from AI products, over 450 MCP clients, and a 13-fold rise in call volume in a quarter, per its July 2026 statements.
What should I ask a data provider before connecting agents to it?
Whether every agent call is checked against your licence with a visible log, whether the licence forbids using the data to train models, and how pricing works at ten times today's call volume.
Does Bloomberg have an MCP server?
Yes. Its Enterprise MCP launched on 29 September 2026 and lets clients' own agents retrieve licensed data with entitlement checks and audit trails, but Bloomberg has disclosed no adoption or pricing.
Disclosure: lowob has no commercial relationship with any provider on this list. Facts checked 17 Oct 2026; next review after the next earnings round, February 2027.