How Mercury Built AI Banking Agents and a Claude-Powered Second Brain
Most startups struggle with financial visibility - manually checking accounts, missing tax savings, and lacking real-time insights. Mercury's banking AI agents solve this by bringing financial data directly into Claude and Anthropic, with some users saving over $1,000 monthly. Discover how their product lead also built a second brain with 5M words of institutional knowledge.
The Evolution of Banking Interfaces
Banking interfaces have transformed dramatically across generations. In the 1950s, customers interacted through physical branches. The 1970s introduced ATMs, while the 1990s brought online banking websites. The 2010s saw mobile apps become the primary touchpoint.
Mercury believes the 2020s will be defined by APIs and AI agents - what Product Lead Ryan Wiggins calls "the conversational interfaces of the future." This philosophy drives their Mercury Connector Platform (MCP), which brings banking functionality into AI tools like Claude.
Key insight: Banking must meet customers where they are - whether that's branches, apps, or now AI agents. Mercury's approach mirrors how Apple transformed computing by creating seamless experiences across devices.
Mercury's MCP: Banking Through AI Agents
The Mercury Connector Platform (MCP) represents their vision for next-generation banking. Available in the Claude app store, it allows users to ask natural language questions about their finances directly within AI interfaces.
As demonstrated at 4:32 in the video, users can ask questions like "How could I save money based on my spending last month?" The system connects to Mercury's API (in read-only mode for security), analyzes transaction data, and provides actionable insights through Claude's interface.
Implementation note: Mercury designed MCP authentication to mirror their web app experience - one-click OAuth with multi-factor authentication. This eliminates the friction of API key management while maintaining bank-level security.
$1,000+ Monthly Savings in Real-World Use Cases
Since launching six months ago, Mercury's MCP has delivered surprising value to businesses. One animation studio discovered tax breaks they didn't know about, while others have optimized recurring expenses.
The company maintains a running tally of customer savings identified through the platform. At the time of recording (12:45 in the video), this number had grown significantly with multiple users reporting over $1,000 in monthly savings from insights provided by their banking AI agent.
Pro tip: The most valuable queries combine financial data with business context. Asking "Where can I save on taxes?" often yields better results than generic spending questions because it leverages Mercury's access to EINs, business addresses, and other identity data.
API vs MCP: Balancing Access and Security
Mercury offers two technical approaches to accessing banking data: their full API and the MCP. The API provides read/write access matching web/mobile functionality, while the MCP version available in AI tools is read-only.
This distinction addresses a key concern at 9:20 in the discussion - ensuring customers maintain control over financial actions while still benefiting from AI insights. The MCP's design prevents scenarios like accidentally canceling subscriptions through conversational interfaces.
Security first: All MCP connections use OAuth with multi-factor authentication, matching Mercury's web security standards. The read-only limitation provides an additional safeguard for financial data accessed through third-party AI platforms.
Building a Claude-Powered Second Brain
Beyond banking products, Ryan Wiggins has implemented an innovative knowledge management system using Claude Code. This "second brain" indexes nearly 5 million words of Mercury's institutional knowledge - including strategy docs, specs, queries, and meeting notes from his five years at the company.
The system (demonstrated at 18:30) uses QMD for local indexing and Claude hooks to inject relevant context into every query. When Ryan asks about activation trends, the system pulls historical data, past analyses, and related discussions to provide answers grounded in full organizational memory.
Key architecture: 1) Local knowledge base (text files), 2) QMD indexing, 3) Claude hooks for context injection, 4) Connected tools (Metabase, Linear, etc.), 5) Multi-agent workflows for complex analysis.
AI as a Performance Coach
One surprising application of Ryan's second brain is performance coaching. By feeding meeting transcripts and feedback into the system, it provides real-time suggestions aligned with his development goals.
At 22:45, Ryan explains how the system catches behaviors like "jumping to solutions too quickly" - a noted area for improvement in his reviews. This creates a feedback loop much faster than traditional quarterly reviews, with his manager reporting noticeable behavioral changes.
Behavioral impact: Knowing the system will flag performance-related patterns has made Ryan more mindful in meetings. The constant gentle nudges are more effective than occasional review discussions for driving lasting change.
How AI Changed Product Development at Mercury
The team's approach to product development has evolved significantly with these AI tools. Three key changes emerged at 27:10 in the discussion:
- Disposable prototypes replaced specs - PMs can quickly edit Mercury's demo environment (demo.mercury.com) to test concepts
- AI data analysts handle 80-90% of routine data questions, freeing human analysts for complex work
- Self-improving systems analyze what questions are being asked to proactively improve data infrastructure
Ryan notes this has accelerated Mercury's product velocity while creating more engaging work for product teams - less documentation, more building.
Future Plans: CLI Tools and Beyond
While committed to MCPs as a distribution channel, Mercury recognizes the need for alternative interfaces. At 16:20, Ryan shares that they're developing a CLI tool for power users who want to optimize context usage.
This reflects their philosophy of meeting different customer needs through multiple access points - whether that's their beautiful web app, mobile experience, API, MCP, or upcoming CLI. The goal remains consistent: make banking data available wherever customers work.
Coming soon: Mercury CLI will provide developers with more direct access to banking data outside conversational AI interfaces, expected to launch in the coming weeks.
Watch the Full Tutorial
See Mercury's MCP in action and Ryan's second brain demo between 4:30-7:45 in the video. The full discussion covers additional insights about measuring MCP success, adoption trends, and how AI is changing product management roles.
Key Takeaways
Mercury's approach demonstrates how established companies can embrace AI while maintaining their core value proposition. By viewing AI agents as the latest evolution of customer interfaces rather than a threat, they've created innovative products that complement their existing offerings.
In summary: 1) Banking AI agents help startups save $1,000+ monthly, 2) Claude-powered knowledge systems can index 5M+ words of institutional memory, 3) The best implementations combine AI with human judgment, 4) Multiple access points (APIs, MCPs, CLI) cater to different user needs.
Frequently Asked Questions
Common questions about this topic
Mercury's MCP (Mercury Connector Platform) is an AI banking agent that connects to Claude and other AI platforms. It allows businesses to ask natural language questions about their finances and get insights.
The system has helped companies identify tax savings opportunities they didn't know about, with some users saving over $1,000 monthly through these insights.
- Read-only access ensures security
- Integrates with existing workflows
- Provides actionable financial insights
Mercury offers both a full API and their MCP. The API provides read and write access matching their web/mobile apps functionality, while the MCP version available in Claude and Anthropic is read-only for security.
This allows businesses to safely access their financial data through conversational AI interfaces without risking unauthorized transactions.
- API: Full read/write access
- MCP: Read-only for AI platforms
- Same authentication standards
Mercury views banking interfaces evolving through history - from branches in the 1950s to ATMs in the 70s, websites in the 90s, mobile apps in the 2010s, and now APIs/AI agents in the 2020s.
Their philosophy is to be available wherever customers are rather than forcing them into a single interface. This has led to innovative products like their MCP that integrate banking into existing workflows.
- Follows interface evolution
- Meets customers where they are
- Balances innovation with security
Ryan built a second brain system using Claude Code that indexes nearly 5 million words of Mercury's institutional knowledge - including strategy docs, specs, queries, and meeting notes from his 5 years at the company.
This system helps him answer questions with full historical context, track performance feedback in real-time, and automate daily workflows like meeting summaries.
- Local knowledge base with QMD indexing
- Claude hooks inject context
- Connected to productivity tools
Mercury's internal AI data analyst handles 80-90% of common data questions from cross-functional teams, freeing up human analysts for more complex work.
The system also provides valuable insights about what questions are being asked most frequently, helping Mercury improve their data infrastructure to better meet employee needs.
- Handles routine data requests
- Identifies knowledge gaps
- Improves over time
Mercury tracks standard product metrics like activation, retention, and expansion for their MCP. They've found strong engagement after setup, with many users incorporating it into weekly or monthly financial review workflows.
The company also maintains a running tally of customer savings identified through the platform, which has grown significantly since launch.
- Standard product metrics
- Usage frequency patterns
- Customer savings quantified
Mercury plans to launch a CLI tool in the coming weeks, providing developers with more direct access to their banking data.
This complements their existing MCP offering by catering to power users who want to optimize context usage and build custom workflows outside of conversational AI interfaces.
- CLI for developers
- Context optimization
- Alternative to conversational UI
GrowwStacks helps businesses implement automation workflows, AI integrations, and scalable systems tailored to their operations.
Whether you need a custom workflow, AI automation, or a full multi-platform automation system, the GrowwStacks team can design, build, and deploy a solution that fits your exact requirements.
- Custom automation workflows
- AI agent implementation
- Free consultation available
Ready to Build Your AI Banking Agent or Knowledge System?
Manual financial tracking costs startups thousands in missed savings opportunities each month. GrowwStacks can implement Mercury-style AI agents or Claude-powered knowledge systems in weeks, not months.