Could a Claude Code routine watch my finances?

Access to banking data (Plaid & alternatives)

• Multiple comments discuss how individuals can or cannot use Plaid. Historically, sales channels required you to be a business with contracts and security reviews.
• A new “hobbyist” / free-trial onboarding flow is mentioned as recently launched, aimed at non-business users.
• Some report still having trouble getting such access in the recent past; timing and availability are somewhat unclear.
• Alternatives mentioned: Yodlee, SimpleFIN (via SimpleFIN Bridge), Tiller (Plaid under the hood), Lunch Money, Actual Budget, Redbark (Australia CDR), Era Finance’s MCP, and direct bank APIs like Monzo’s.

Using LLMs and agents for personal finance

• Several users describe building their own pipelines: e.g., bank → Tiller/CSV → Google Sheets → Supabase/SQLite → MCP → LLM for analysis.
• LLMs are used for transaction categorization, subscription detection, cashflow projections (including Monte Carlo simulations), and scenario questions like “what mortgage can I afford?”.
• Some see strong potential for LLM-based financial advisors, always-on agents, and email/chat-style “ambient” advisors that monitor and notify instead of gamified dashboards.
• Others note companies have already tried and failed economically in this space.

Security, privacy, and trust concerns

• Deep unease about giving Plaid bank credentials and about data sharing/sale risks.
• Debate over whether Plaid still stores credentials vs using OAuth; consensus: big banks mostly use OAuth/tokenized APIs, smaller ones often still involve scraping and stored credentials.
• Some argue the correct stance is to avoid such automation until proper open banking APIs exist.
• The showcased product emphasizes read-only Plaid scopes, no money-movement tools, segregated encrypted tokens, strict IAM, and narrow MCP tools.
• There are warnings about routine/agent modes: all tools (including write-capable ones in general) may be callable, so prompt-injection and “rogue agent” risks must be considered.

Perceived value vs traditional tools

• Enthusiasts praise the flexibility of getting data out of closed banking apps and into systems they control, and the ease of wiring multiple data sources together.
• Skeptics say spreadsheets + deterministic tools (Tiller, Actual Budget, plain-text accounting) are enough and more trustworthy, especially given LLM hallucinations with financial data.
• Some worry about obsessive tracking and daily net-worth checks; prefer infrequent, calm financial reviews.
• A subset wants self-hosted, open-source, or strictly local-LLM solutions due to strong privacy norms.