Inside Grasshopper’s First-of-Its-Kind MCP-Powered Offering for SMBs
By Nicole Volpe, Contributor at The Financial Brand
Simple Subscribe
Subscribe Now!
Executive Summary
- Grasshopper Bank became the first U.S. community financial institution to deploy a Model Context Protocol server, enabling business clients to query their banking data using natural language through AI assistants like Claude.
- The initiative tackles a longstanding market problem where small and medium businesses have been underserved by digital banking, often forced to use consumer-grade platforms not designed for business needs like B2B payments and liquidity management.
- While the current implementation only allows data querying and analysis for security reasons, future updates will gradually introduce limited write capabilities like editing transaction categories and adding notes, with broader bi-directional features still years away.
Grasshopper Bank, a New York City-based all-digital bank with $1.4 billion in assets, has rolled out a new AI-driven capability that aims to redefine how its client base — small businesses and startups — access and interact with their financial data. At the heart of Grasshopper’s new user experience is a first-of-its-kind Model Context Protocol (MCP) server, which enables the bank’s business banking clients to securely query accounts using natural language via AI assistants like Anthropic’s Claude.
Grasshopper, which focuses primarily on small and medium-sized businesses (SMBs) and startups, implemented its MCP in partnership with Narmi, a tech firm that helps community banks and credit unions better leverage their digital capabilities. The bank’s MCP integration is in beta test mode with a limited group of clients, with plans to expand across its wider client portfolio later in 2025. The deployment is considered the first natural language interface to be operationalized for clients by a community financial institution in the U.S.
In tackling the project, the bank was responding to what it had identified as a long-standing market gap: SMBs have historically been underserved by digital banking, often forced to make do with consumer-grade platforms that aren’t specifically designed to support businesses’ needs, like B2B payments or liquidity management.
The effort is further anchored by the belief that user experience can in fact be a primary lever for banking company growth, and that SMBs can gain a meaningful competitive advantage by reducing friction in everyday financial tasks. “I think about things in simple terms — time and friction,” said Mike Butler, Grasshopper Bank’s CEO. “Reducing the friction of working with a financial services company for a small business is the number one experience that we’re constantly focused on.”
The bank and the tech firm focused on building a platform that mirrors how business owners actually work: intuitive discovery of insights, streamlined payments, and integrations with the tools companies already use.
“Businesses are not going to stay in a bank’s neatly walled-off garden; it’s impossible,” Narmi Co-Founder Chris Griffin said. “They’re going to use other tools and our goal is make those other tools accessible through the bank’s MCP.” This approach also has the potential to help community banks better compete with — and even leapfrog — fintech competitors whose apps and SaaS tools have long held a customer experience edge.
Want to read more like this? Check out Narmi’s content portal on The Financial Brand: Be Where Banking is Going
The MCP Connection
Grasshopper’s integration employs MCP — an emerging standard that makes it possible for enterprises to connect sources of data directly to large language models — to enable its business banking clients to pull their financial data into Anthropic’s Claude.
The initiative positions conversational AI as a core feature of business banking and financial analysis. Instead of exporting spreadsheets or toggling between multiple dashboards — often amid cluttered and hard-to-navigate user interfaces — business owners or their finance teams can now ask questions and get real-time answers based on the company’s live transaction data.
At the basic level, a Grasshopper client opens Claude, authorizes a secure connection to their Grasshopper data, and then asks plain-English questions about active balances, expenses, cash flows, and more — e.g, “How much did my company spend on software last month?” The AI assistant will return insights, suggestions, and visualizations — all of its native capabilities — based on trends in this data.
A business owner might spot-check cash-flow trends to quickly identify duplicate charges or expense-category outliers, without the need to export CSVs or build spreadsheets. “MCP is giving them a secure way to do things that they’re going to do anyway in much clunkier, riskier ways,” Griffin said.
The Grasshopper implementation draws data not just from the bank’s core system but also, critically, from permissioned third-party sources like card processors and bill pay platforms. “Banks often think about their digital banking platform as a portal or a destination, but that’s not how businesses are actually working today,” Griffin said. “They’re in multiple systems all the time, whether it’s their accounting platform, a payments tool, or now increasingly an AI assistant.”
This opens the door to more sophisticated financial analysis and complex tasks. So a startup whose finance head might already have been using AI to build board decks now has everything in one place, with no need to manually search, download, and enter bank or accounting system data, since analyses and presentations can be generated directly within an AI session. They can generate projections and what-if scenarios: e.g., “Show me the impact on cost of sales if two additional $100,000 FTEs are included plus commissions and referral fees.” Armed with MCP-enabled accounts, an SMB with an outsourced CFO function can request extended analyses like these without incurring large additional fees.
Though the MCP spec can handle read/write access, to limit compliance and data security risk, Narmi and Grasshopper chose to keep the implementation read-only for now: Clients can query and analyze their data but cannot initiate payments or change account settings through the AI.
Integration and Extension
The MCP rollout is one component of the larger implementation by Grasshopper of Narmi’s digital business banking platform. The platform introduced a set of key new client-facing features for the SMB market, including digital money movement (account transfers, bill pay, wire and ACH) and administrative controls such as role-based permissions and dual approvals so customers can internally delegate work while maintaining oversight.
Grasshopper was able to accelerate the platform rollout by leveraging Narmi’s existing integration of FIS’s IBS core system, which the bank had adopted for digital account opening. Building on that connection — along with Narmi’s open platform and API framework — the bank completed the business banking platform rollout in under five months and with minimal time investment by its own staff.
Narmi’s open API also allowed Grasshopper to embed widely used third-party services directly into its platform. Available integrations included QuickBooks and Autobooks for accounting and invoicing, Alloy for fraud prevention, and MX for data aggregation and financial insights. Grasshopper CEO Butler described the goal as creating an “Amazon-like” environment in which business clients could access essential services in one place rather than managing multiple disconnected systems. The framework also supports fast expansion in the future, he said.
Butler said: “We need to think about what a small business does every day…. What takes time, and how we can reduce friction in that experience. How many questions does a client really need to answer? How many clicks and how long does it take? How do you navigate the website? Does it make sense? Can you track your wire transfers?”
Looking ahead, Grasshopper intends to continue its development work with Narmi, to bring more innovative services to market that are better integrated in clients’ operational needs. In addition to functional improvement releases to keep the platform current, delivered on a six-week cadence, Narmi updates have to date included a range of design and usability changes, such as client login redesign.
Updates to the MCP server deployment are also in the pipeline, including taking better advantage of the still untapped read/write capability. According to Narmi’s Griffin the first potential expansion would be narrow: allowing users to write metadata back to their records — for example, changing a transaction category, editing a description, adding a note, or attaching a document.
Griffin emphasized that this incremental approach reflects both security considerations and the current state of generative AI. More extensive write capabilities, such as initiating transfers or updating account information, would introduce compliance risks that the industry has not yet fully addressed. Broader bi-directional use is therefore “pretty far off,” as Griffin put it, and will only be reconsidered once the technology matures and banks are confident in its reliability.
