What is the Model Context Protocol (MCP)?

The Model Context Protocol (MCP) is an open standard that lets AI assistants — Claude, ChatGPT, Gemini, and others — connect directly to external data sources and tools. Instead of copying information into a chat window, MCP gives your AI secure, structured access to live data. Era is one of the first platforms to use MCP for personal finance, turning your bank accounts, credit cards, and investments into a data layer that any AI you already use can read and act on.

If you've ever wished you could just ask your AI "what did I spend on groceries last month?" and get a real answer from real data, MCP is the technology that makes that possible.

The problem MCP solves

AI assistants are remarkably good at reasoning, planning, and explaining. But they have a fundamental limitation: they can only work with information you give them. Ask Claude about your finances, and it can offer generic budgeting advice. It can't tell you that your electricity bill jumped 40% this month, because it can't see your electricity bill.

Historically, people solved this problem in clunky ways:

• Copy-pasting: Export a CSV from your bank, paste it into the chat, hope the AI parses the columns correctly.
• Screen scraping: Automated tools that log into your bank and pull data — fragile, often against terms of service, and a security nightmare.
• Proprietary chatbots: Finance apps that build their own AI chatbot inside their app. You get AI, but only their AI, only inside their app, using only their model.

Each approach has the same structural flaw: the data is trapped. Either it's trapped in a file format your AI has to guess at, trapped behind a scraper that breaks when the bank changes a button, or trapped inside an app that chose your AI for you.

MCP eliminates the trap.

How MCP works, simply

MCP defines a standard way for AI clients (the apps you chat with) to connect to MCP servers (the systems that hold your data). Think of it like USB for AI — a universal connector that lets any compatible device talk to any compatible peripheral.

An MCP server exposes tools — structured actions the AI can call. These are not free-text prompts. They're typed, documented endpoints: "list my bank accounts", "search transactions from last week", "analyze spending by category". The AI reads the tool descriptions, decides which ones to call based on your question, and presents the results in natural language.

The key properties:

• Open standard: Anyone can build an MCP server. Anyone can build an MCP client. No single company controls the protocol.
• Structured data: The AI receives typed, clean data — not a blob of text it has to parse. This means better, more accurate answers.
• Permissioned access: You decide which MCP servers your AI can connect to. You can revoke access at any time.
• Client-agnostic: One MCP server works with every MCP-compatible client. Build once, connect everywhere.

What this means for personal finance

Personal finance is one of the best use cases for MCP. Your financial data is deeply personal, constantly changing, and incredibly useful for the kind of reasoning AI is good at — spotting patterns, forecasting, comparing, planning.

Before MCP, getting AI to work with your money meant one of two things:

1. Upload your data manually every time you want an answer. Tedious, error-prone, and the AI forgets everything between sessions.
2. Use a finance app's built-in chatbot, which locks you into that app's AI model, that app's interface, and that app's idea of what questions you're allowed to ask.

MCP creates a third option: your financial data becomes a persistent, secure layer that any AI you choose can access. You're not locked into any particular AI client. You're not uploading files. Your data stays with your data provider, and your AI queries it live.

How Era uses MCP

Era Context is a personal MCP server for your finances. You connect your bank accounts to Era through MX (a regulated financial data provider), and Era exposes that data as a set of 33 MCP tools across seven groups: accounts, transactions, insights, activity, billing, knowledge, and connections.

Setting it up takes one line of configuration. In Claude Desktop, Cursor, VS Code, or any other MCP-compatible client, you point to https://context.era.app with your authentication token. That's it. Your AI can now see your money.

Here's what becomes possible:

• "What's my checking account balance?" — answered from live data, not memory.
• "How does my restaurant spending this month compare to last month?" — computed from your actual transactions.
• "Show me all recurring charges over $50" — pulled from pattern analysis across your accounts.
• "Remember that I'm saving for a trip to Japan" — stored in Era's cross-agent memory, so every AI you connect knows about your goal.

That last point is worth dwelling on. Era's cross-agent memory means that if you tell Claude about your savings goal, ChatGPT knows about it too. Your financial context persists across conversations and across AI clients. No re-explaining.

Beyond querying: memory and automation

MCP is not just about reading data. Because MCP tools can accept inputs as well as return outputs, an MCP server can offer write operations — actions your AI can take on your behalf.

Era uses this to enable two capabilities that fundamentally change how you manage money:

Cross-agent memory. Era Context includes a knowledge system that persists across conversations and across AI clients. Tell Claude that you're saving $500 a month for a trip to Japan. Later, open ChatGPT and ask "am I on track for my savings goal?" ChatGPT already knows about the trip, the target amount, and your progress — because Era stored that context, not any individual AI client. You can also ask any agent to forget something, and it's gone everywhere. Your memory is private to you, never shared with other users, and never used to train models.

Plain-English automation rules. Describe a rule to any connected AI — "categorize all Starbucks transactions as coffee" or "tag any charge over $500 as worth reviewing" — and your AI creates it through Era's rules engine. You approve before it activates. Nothing runs without your explicit sign-off. Every rule remembers the exact words you used to create it, giving you a full audit trail in your own language. A library of pre-built rules is also available to browse and activate without writing anything from scratch.

These capabilities are only possible because MCP provides a standard, structured way for AI to interact with external systems. Without MCP, each AI client would need its own custom integration — and your data would be siloed in whichever client you happened to use.

MCP vs. the walled-garden chatbot

Most finance apps that offer AI follow the same playbook: embed a chatbot inside the app. You open the app, tap the chat icon, and talk to whatever model they chose for you.

This approach has real limitations:

• Model lock-in: You use their model, not yours. If a better model launches tomorrow, you can't switch.
• Interface lock-in: You have to be inside their app. You can't ask about your finances from Claude Desktop, or from VS Code while you're coding, or from whatever AI tool you use throughout your day.
• Context isolation: The chatbot only knows what's inside that app. It can't connect your financial data with your calendar, your email, your project management tools, or anything else your AI has access to.

MCP servers flip this model. Era doesn't have a chatbot. Era has a data layer. Your AI of choice is the interface. This means you can ask about your finances wherever you already are — in the middle of a conversation about trip planning, while reviewing a contract, while building a budget spreadsheet. Your money shows up in context, not in a separate app.

Security and trust

Connecting your bank accounts to AI raises legitimate security questions. MCP addresses several by design:

• No credential sharing: Your bank login credentials are handled by MX during authentication and are never stored by Era or seen by your AI.
• Scoped access: Each MCP tool has defined inputs and outputs. Your AI can call "list transactions" but can't access raw database tables or internal systems.
• Revocable: You can disconnect any AI client from your Era Context at any time, instantly cutting off access.

Era adds additional protections on top of MCP:

• AES-256 encryption at rest, TLS 1.3 in transit.
• Your data is never sold, never used for advertising, and never shared without your explicit permission.
• Every AI agent interaction requires explicit authorization.
• A full activity log shows everything any AI agent has done on your behalf.

Which AI clients support MCP

MCP adoption is growing rapidly. Era works with any MCP-compatible client, including Claude, ChatGPT, Cursor, VS Code, GitHub Copilot, Gemini, Perplexity, OpenClaw, Manus, Cline, Hermes, and more. The list expands regularly as more AI tools adopt the standard.

The important point is not any specific client — it's that you're not locked in. When a new AI client launches and supports MCP, it works with Era on day one. No integration needed, no feature request, no waiting. One protocol, every client.

Why MCP matters more than any single AI model

AI models improve rapidly. The model you use today will likely be surpassed within months. This creates a problem for any platform that couples its AI features to a specific model: every model upgrade becomes a migration, and users are stuck with whatever the platform ships until the next update.

MCP decouples the data layer from the AI layer. Your financial data is accessible through a stable protocol regardless of which model is on the other end. When a new model launches — faster, cheaper, better at reasoning — you connect it to the same MCP server and get immediate benefits. No migration, no waiting for your finance app to integrate it, no re-training a chatbot.

This is why Era built on MCP rather than embedding a specific model. The protocol outlasts any individual model generation. Your data stays structured and accessible, and the best AI available at any given moment can work with it.

Getting started

Era's Basic tier is free and includes two connected accounts with read-only MCP access. You can sign up at era.app, connect a bank account, add one line of configuration to your AI client, and start asking questions about your money in natural language.

The setup takes about five minutes. The shift in how you think about your finances takes a bit longer — but once your AI can actually see your money, you'll wonder why you ever managed it any other way.