MCP for Financial Services: Market Data Connectors, FINRA Expectations and Controls

MCP lets an AI assistant pull fundamentals, estimates, index data and transcripts from licensed market data providers, and query internal systems, from one conversation. Research analysts, bankers and operations teams are the early users. For regulated firms, existing rules on supervision, communications and customer data apply to these tools just as they apply to people.

MCP (Model Context Protocol) is an open standard for connecting AI apps to other software through MCP servers. If you need the basics, read What Is MCP?.

This guide covers what MCP can do in financial services, which servers are documented, what FINRA and the SEC have said, and a starter stack. Facts were checked on 2 November 2026. It is not legal or compliance advice: your chief compliance officer and counsel decide how these rules apply to your firm. If you mainly need bookkeeping, read MCP for Accounting and Bookkeeping instead.

What MCP can do in financial services

Good first uses are research and preparation, not execution:

• Company research. Pull fundamentals, estimates and recent transcripts for a company and draft a tear sheet.
• Index and exposure questions. Look up index constituents, performance or exposures for a portfolio review.
• Meeting prep. Combine market data with CRM notes to brief an adviser before a client call.
• Operations diagnostics. Investigate a failed bank link or a funnel drop in a fintech app.
• Internal Q&A. Answer staff questions from policy manuals and procedures through a read-only server.

Each output is a draft for a person who is accountable for it.

Which financial services MCP servers exist

FactSet

FactSet's AI-Ready Data MCP page describes one server that aggregates its content APIs. It lists estimates, pricing, fundamentals, ownership, events and transcripts, supply chain relationships and more "across 116 countries and 86,000+ companies." The page shows 23 tools and the endpoint https://mcp.factset.com/content/v1. It lists clients including Claude, ChatGPT, Microsoft Copilot, Gemini and Glean. Ask FactSet about access terms for your firm.

Connectors named by Anthropic

Anthropic's February 2026 finance plugins post names:

• FactSet: an MCP connector for market data, fundamentals, estimates and research.
• MSCI: an MCP connector for index data, including performance, exposures, constituents and methodologies.
• S&P Global: a partner plugin that brings Capital IQ Pro into Claude.
• LSEG: a partner plugin for LSEG data and analytics, "Available for users with active LSEG data entitlements."

These connectors carry your existing data licenses into the AI tool. They do not grant new rights.

Fintech and payments servers

• Plaid. Plaid's Dashboard MCP server provides "Production diagnostics and analytics tools," such as debugging a linked account connection and Link conversion metrics. It "works with Production data only." See our Plaid MCP server profile.
• Stripe. Useful for payment, invoice and payout data. See the Stripe MCP server profile.
• Bookkeeping. For the general ledger side, compare options in Best MCP Accounting and Bookkeeping Tools. That list includes LedgerMCP (built by eSEOspace).

Core banking and custody

We did not find core banking, custody or order management vendor MCP documentation we could open and verify on 2 November 2026. Those systems are often where firms build their own servers, with strict read-only scopes.

What regulators have said

FINRA

Regulatory Notice 24-09, dated 27 June 2024, reminds member firms of their obligations when using generative AI. It says "FINRA intends for its rules and guidance to be technologically neutral." Points that map to MCP:

• Supervision (Rule 3110). A firm "must have a reasonably designed supervisory system tailored to its business." If AI is part of supervision, policies should address "technology governance, including model risk management."
• Communications (Rule 2210). Content standards apply "whether" a communication is generated by a human or a technology tool.
• Vendors. Obligations apply to third-party technology, "including through embedded features in existing third-party products." Firms should "evaluate Gen AI tools prior to deploying them."

The notice says it "does not create new legal or regulatory requirements."

FINRA's January 2026 post on AI agents is closer to what MCP enables. It lists risks including "AI agents acting autonomously without human validation and approval," agents that "act beyond the user's actual or intended scope and authority," multi-step reasoning that complicates auditability, and agents that may "store, explore, disclose or misuse sensitive or proprietary information." Those map directly onto MCP controls: approvals, scopes, logging and data minimization.

SEC Regulation S-P

The SEC adopted amendments to Regulation S-P on 16 May 2024. Covered institutions must maintain "written policies and procedures for an incident response program" for unauthorized access to customer information, and notify affected individuals "as soon as practicable, but not later than 30 days" after becoming aware of an incident. Larger entities had 18 months and smaller entities 24 months after Federal Register publication to comply.

Every MCP server that can read customer information belongs in that incident response scope. Plan how you would revoke a connector's access in minutes if something goes wrong.

Recordkeeping and logging

Ask your compliance team which AI interactions count as books and records for your firm. Whatever the answer, you will want logs of each tool call: who, which server, which tool, what inputs and when. Keep those logs for as long as your retention policy requires.

Example prompts

• "Pull the last eight quarters of revenue and EPS estimates for these three companies and build a comparison table."
• "List the top ten constituents of this index and their weights, with the data source for each figure."
• "Summarize the last earnings call transcript for this company in five bullet points, with quotes."
• "Why did this user's bank link fail yesterday? Show me what the diagnostics tool returns."

Ask for sources in every prompt. Analysts and supervisors can then check the numbers against the vendor's own terminal.

Questions for your compliance team

1. Which AI interactions count as books and records for us?
2. Which communications drafted by AI need principal approval?
3. Which vendors may receive customer information, and under which agreements?
4. How will supervisors review agent activity?

A starter stack

Use our tool permission planner to mark each tool allow, ask or block.

What to keep human

• Investment recommendations and suitability judgments.
• Order entry and any movement of money.
• Communications to customers and the public, which may need principal approval.
• Supervisory reviews and exception handling.
• Account opening and identity decisions.
• Decisions about which vendors receive customer data.

The MCP specification says there "SHOULD always be a human in the loop with the ability to deny tool invocations," per the 2026-07-28 tools specification.

Next step

Our MCP readiness audit maps your data, entitlements and supervisory needs before you connect an AI. If you need a custom read-only server over internal systems, see MCP development. For a side-by-side list, read Best MCP Servers for Finance Teams.

Frequently asked questions