Visa Put an MCP Server in Its Annual Report — And Finance Is About to Change Forever
When a company files its annual report with the SEC, every word is reviewed by lawyers, auditors, and executives who know it becomes a legal document. Nothing makes it into a 10-K by accident.
So when Visa — the company that processes over $15 trillion in annual payment volume — disclosed an MCP server in its FY2025 annual filing, that wasn't a product announcement. It was something more significant: an audited fact about how the world's largest payment network is connecting to the AI agent stack.
What Visa Actually Said
Visa's FY2025 10-K (Item 1, filed November 2025) describes how its network is being made accessible to developers and AI systems. The relevant passage reads:
"to provide value by enabling access to our global network through multiple integration methods, including programmatic access via application programming interfaces (APIs) and structured data exchange through our Model Context Protocol (MCP) server."
And further in the same filing:
"These capabilities are supported by a commercial partner program for platforms and developers, providing the tools and infrastructure to deploy AI-driven commerce solutions securely and at scale. Additionally, these APIs are accessible through our MCP server, connecting AI systems to Visa's payment network infrastructure and Visa's Trusted Agent Protocol, which enables safer agent-driven checkout."
This isn't a blog post or a keynote. This is the Item 1 Business section of Visa's 10-K — the part where management describes what their business does. Visa is telling the SEC, its shareholders, and the world: MCP is now part of how we distribute our network.
The language matters. "Structured data exchange through our Model Context Protocol server" — that's a technical architecture disclosure. Visa is describing MCP not as an experiment but as an established integration method, placed in the same sentence as APIs, a technology the company has used for more than a decade.
The "Trusted Agent Protocol" is the more significant piece. This is Visa's framework for ensuring that when an AI agent initiates a payment — not a human clicking a button — the transaction can still be authenticated and the liability framework holds. Credit card networks depend on clear chains of authorization; the Trusted Agent Protocol is Visa's answer to the question "who is responsible when an AI agent pays for something on your behalf?" This is infrastructure-level work, not a demo.
CrowdStrike Sees the Same Trend — From the Security Side
While Visa is building MCP infrastructure, CrowdStrike is selling security for MCP infrastructure. CrowdStrike's FY2026 10-K (Item 1, filed March 2026) describes a product category that wouldn't exist unless enterprise MCP deployments were already widespread enough to need protecting:
"AI Detection and Response ('AIDR') provides visibility and governance into how employees use AI and how AI agents operate by mapping relationships between users, prompts, models, agents, and Model Context Protocol ('MCP') servers, and enforcing policy across these relationships."
CrowdStrike isn't launching an MCP server. They're selling governance tooling for MCP server deployments — which means their enterprise customers are running enough of them that mapping, monitoring, and policy enforcement is a commercial product category.
This is how you read the maturity curve of any infrastructure technology: the security vendors arrive slightly after the builders. When the leading enterprise security vendor ships a product line specifically around protecting MCP server deployments, you're no longer in "early adopter" territory. You're in "this is infrastructure and it has attack surfaces" territory.
Stripe Was Already First
Before either of those filings, Stripe launched its official MCP server in 2025. It exposes billing, subscription, customer, and payment intent data to AI agents with full authentication. Claude and other AI systems can call Stripe tools to look up customer payment history, create invoices, check subscription status, and retrieve receipts — directly, without a human in the loop.
EvidInvest is already connected to Stripe's MCP server. When an AI agent needs to check billing status or trigger a subscription action, it calls Stripe tools directly. This isn't theoretical — it's how our billing automation works today.
Stripe's implementation established the pattern that Visa is now scaling: define a clean tool surface, authenticate at the tool level rather than the user session level, and let AI agents do what they'd otherwise require a developer to build custom integrations for.
What the Visa MCP Opportunity Looks Like
Visa's disclosure indicates its MCP server is currently in a "commercial partner program" phase — accessible to platforms and developers, not yet publicly available to every Claude user. Based on the filing's language, here's what it likely exposes to AI agents:
- Transaction data queries — "Did this merchant charge me twice?" "Show me all Visa transactions above $500 this quarter."
- Merchant analytics — Average spend per customer at a merchant category, frequency patterns, geographic distribution.
- Network intelligence — Fraud rate signals, BIN range data, authorization decline patterns.
- Tokenized payment execution — Executing a purchase on behalf of an authenticated user, with the Trusted Agent Protocol providing the authorization trail.
The Trusted Agent Protocol is the key innovation enabling that last capability. Visa is building the authentication layer so AI agents can transact, not just query. The first commercial use case is probably agentic checkout: an AI books your flight, pays with your Visa card, with cryptographic proof that you authorized the agent to act on your behalf. The enterprise use cases — automated procurement, recurring payment management, spend analytics pipelines — go further.
For AI builders, this matters because the hardest part of building financial AI agents isn't the language model — it's the data access and payment execution layer. Visa's MCP server is solving that second layer. The first layer — financial data and valuation intelligence — is what EvidInvest solves.
The Agentic Finance Stack
We're at the beginning of an agentic finance layer being assembled piece by piece. Each major participant in the financial system is staking out its MCP position:
| Layer | What It Does | MCP Provider |
|---|---|---|
| Financial data & valuation | DCF models, fair value, EPS trajectory, screeners | EvidInvest (46 tools) |
| Payments & billing | Subscriptions, invoices, payment history | Stripe (live) |
| Payment network | Transaction execution, merchant data | Visa (partner program) |
| Portfolio & brokerage | Trades, holdings, P&L | Emerging (IBKR, Schwab TBD) |
| Banking | Account balances, transfers | Plaid-style (TBD) |
Each layer becoming MCP-native means AI agents can eventually query, analyze, and execute across the full financial lifecycle — in a single context window, without stitching together bespoke API integrations. An agent that can check a company's valuation, review the customer's payment history, execute a trade, and log the transaction against a billing record is not science fiction. It's the natural endpoint of what Visa, Stripe, and EvidInvest are separately building toward.
EvidInvest's position in this stack is the analysis and intelligence layer. Not just prices (every data vendor has prices), not just financial statements (those are commoditized) — but pre-computed DCF models, EPS trajectory classifications, growth rate estimates, and fair value ranges. The analysis that takes an analyst hours, available to any AI agent in one tool call.
When Visa's MCP server is broadly available and an AI agent is executing an equity-linked payment or an automated reinvestment trigger, it will need a valuation layer to decide whether that execution makes sense. That's the integration point.
What This Means for You
If you're building financial AI agents today: these MCP servers are real and available now. Start with EvidInvest for valuation intelligence and Stripe for billing — those are the two best-documented, most production-ready options in the finance space. Watch what Visa publishes as its partner program opens up; the Trusted Agent Protocol will be significant for any agent that needs to transact, not just analyze.
If you're an investor watching this space: MCP adoption in financial services is showing up in 10-Ks, not press releases. Visa's disclosure and CrowdStrike's security product category are the kind of signal that tends to be a few quarters ahead of the market's understanding. That's usually a better indicator than keynote announcements.
The infrastructure layer of agentic finance is being built now. The companies that are MCP-native at the data, billing, and execution layers will have a structural advantage over those that aren't — because the AI systems that financial professionals, developers, and consumers use will simply reach for what's available.
→ Explore EvidInvest's 46 MCP tools at evidinvest.com/developers
EvidInvest provides financial data and valuation tools for educational purposes. Nothing in this post is investment advice.
Fair Value Weekly
Get DCF breakdowns, fair value updates, and portfolio ideas for serious investors. No spam, no paywalled teasers.