How to Build for AI Agents and a Claude Code Second Brain in 25 Min

Wiggins walks through Mercury’s MCP, which lets customers ask natural-language questions about their bank data in tools such as Claude. The connector is read-only, with OAuth and multi-factor authentication, while Mercury’s broader API supports read and write access. One animation studio reportedly used it to find tax breaks, and some users have reported saving more than $1,000.

Wiggins says agent-ready products depend on solid core services and broad API coverage before adding an MCP. Mercury aimed to make connecting feel like signing in to its web app, with one-click OAuth and the same password and multi-factor authentication. He sees MCP as a way to reach users through conversational apps, while a Mercury CLI is planned for developers who want more flexible tools.

Wiggins describes a local knowledge base built from nearly five million words of Mercury documents and work history. Claude Code uses QMD indexing and hooks to bring relevant context into questions, with skills, integrations, and multi-agent workflows for analysis. His daily routine includes briefs drawn from his calendar and work tools, plus end-of-day meeting summaries and action points.

Wiggins says an internal AI data analyst answers an estimated 80 to 90 percent of common cross-functional questions. His second brain also helps him reflect on meeting behavior by comparing transcripts with themes from performance reviews. At Mercury, disposable prototypes help teams explore ideas sooner, and the AI analyst helps reveal which questions need better data infrastructure.

Wiggins describes a recent shift in which Anthropic became the first AI model chosen by more new startup cohorts than OpenAI, based on Mercury’s analysis. He says accumulated context and configured workflows can make switching tools difficult. For product teams, he sees AI accelerating prototyping and access to information, while the PM’s responsibility to align people and ship work remains.