We built an MCP server that gives Claude/ChatGPT access to your actual financial data. Founder here, looking for honest feedback.
👋 Cofounder of Era Finance here. We built an MCP server called Era Context that connects your bank accounts to whatever AI assistant you're already using (Claude, ChatGPT, OpenClaw).
Since agents need context and they don't have it otherwise, we thought this would be the right direction to keep our data private, but allow agents to tap into it.
Anywho - Era Context sits underneath or between the banks and agents. It connects to your accounts, cleans up transaction categorization (bank categories are bad), lets you set custom tags and automation rules, and exposes the right details over MCP so any connected assistant can actually reason about your real numbers. Plus, cross agent memory, tell Claude a goal, and it's still there when you open ChatGPT.
Things I'm genuinely unsure about and want pushback on:
- Is cross-agent memory actually a problem people feel day to day, or is this a problem builders care about more than users do?
- The obvious objection is trust. Handing financial data to an MCP server that then hands it to an LLM. What would actually make you comfortable with that, versus what's just a nice-sounding security page?
- We kept it read-mostly for now (no money movement yet) partly for exactly that trust reason. Curious if that feels like the right tradeoff or overly cautious.
- We have lots on our roadmap, but starting on the basics first.
Not asking for sign ups. Happy to answer anything in the comments, technical or skeptical. Shared the link if people want to poke at it. Can also offer dummy data to test our manual tools.