MCP and the Quiet Death of Fintech APIs as We Know Them
Anthropic's Model Context Protocol launched in late 2024. Most people read it as a developer convenience. Inside fintech, it is the start of a much bigger rewrite.
In November 2024, Anthropic released the Model Context Protocol — an open standard for connecting AI assistants to data sources and tools. Developers welcomed it as a cleaner alternative to writing one-off LangChain integrations. What happened next inside fintech is more interesting: every serious team started asking what their stack looks like if MCP becomes the way AI agents access bank data, market data, and trading systems.
What MCP actually is
Strip the protocol-spec language away and MCP is a contract: a server exposes capabilities (tools, resources, prompts) in a standardized way, and any MCP-aware client (Claude Desktop, Cursor, an internal agent) can discover and use them.
It is conceptually similar to LSP (Language Server Protocol) for code editors. Once the protocol exists, the ecosystem flips: instead of every editor implementing every language and every language integrating with every editor (an N x M problem), you implement the protocol once.
Why this matters for fintech
Fintech is the API-economy poster child. Plaid, Truelayer, Stripe, Yodlee, MX — these are companies whose entire value is being the connective tissue between consumer-facing apps and bank backends. Every one of them exposes REST APIs, has an SDK in 6 languages, charges per call or per connected account, and has documentation that takes a week to ramp on.
Now imagine: every bank, every brokerage, every data provider, every accounting system exposes an MCP server. An AI agent can discover capabilities, authenticate, and execute workflows without anyone writing custom integration code.
Two questions follow:
- What happens to fintech aggregators whose moat is integration breadth?
- What new businesses become possible when an end user's AI agent can directly talk to their bank, broker, and accountant?
The optimistic case for incumbents
Plaid does not lose by MCP existing — they lose if they fail to become the dominant MCP server for banking. Their data, normalization, identity verification, and compliance work are still valuable. The question is whether they package it as MCP-native or keep insisting agents talk to their REST endpoints.
Same for Stripe: a Stripe MCP server that exposes payments, subscriptions, treasury, issuing in a way an agent can reason about is more valuable than a REST API. The early signals from Stripe's Agentic Toolkit (released late 2024) suggest they see this.
The disruption case
The companies that should be worried are the ones whose value is mostly "we built integrations with 12,000 banks." If every bank exposes its own MCP server (which they will, because the alternative is fighting the entire AI ecosystem), the aggregator moat erodes fast.
On the other side, a wave of new fintech products becomes possible:
- True "set-it-and-forget-it" finance agents — your AI handles bill pay, savings allocation, tax optimization, with explicit auditable actions.
- Bookkeeping that closes itself — accounting agents that pull data from every source, categorize, reconcile.
- Compliance copilots — agents that query bank, brokerage, and regulatory systems to generate filings.
- Cross-institutional portfolio analysis — your agent sees everything (held-away assets, real estate, crypto, alternatives) and gives unified advice.
What I am paying attention to
Three things to watch in 2026-2027:
- Which big bank (JPMorgan, Citi, BofA, Goldman) is the first to ship a public MCP server for retail banking. That moment is the inflection point.
- Whether OpenAI standardizes on MCP or forks the protocol. The ecosystem needs the convergence.
- Regulatory response — the SEC and OCC will have opinions about AI agents executing trades or moving money. The early frameworks here will shape the next decade.
Every shift in how software talks to software creates a generation of new companies and kills a generation of incumbents. SOAP to REST. Monoliths to microservices. REST to GraphQL. MCP is the next of these — and fintech, being the most API-dependent industry, is where the shift hits hardest first.
Let's talk
I'm pivoting from manufacturing AI to finance — open to roles, mentorship, and collaborators in fintech, quant, and bank AI.