Connecting an AI assistant to financial data usually means building the interface yourself. For every use case and every client, again and again. That costs time, and it often opens up systems further than the task actually requires.

An open standard for access

There is now an open standard for this: the Model Context Protocol (MCP). MCP defines how an AI client communicates with a data source—via clearly typed, curated tools rather than unfiltered database access. A client calls a limited, defined function; the system behind it returns structured, validated results instead of guessed values.

On this basis, we have released our own MCP server. It is listed in the official MCP registry as eu.wealthapi/mcp-server gelistet, der Quellcode liegt öffentlich auf GitHub, und die Installation erfolgt über npm.

What the server provides

The wealthAPI MCP server provides 21 functions—read-only only. It connects account, investment, transaction, and profile data, as also available via the wealthAPI API. An MCP-compatible client like Claude Desktop can use it to analyze, among other things:

• Portfolio performance over a selected period
• Cash flow, savings rate, and recurring costs
• Dividend history and calendar
• Winners and losers in the holdings

In addition, free-form queries are also possible because the server can access the full transaction history—for example, “Do you see any unnecessary subscriptions?” or “Are these expenses typical for a three-person household?” The answer is not based on an estimate by the language model, but on the data actually retrieved via the tools.

Architecture: local, read-only, limited

Three characteristics are critical for use:

Local instead of hosted. The server runs as an npm package on your own machine. There is no hosted version. The API token does not leave your environment and is not transmitted to wealthAPI servers.

Read-only instead of write access. All 21 tools are limited to read-only access. There is no function that triggers transactions or changes data.

Limited instead of full access. Access is restricted to four scopes (accounts:read, investments:read, transactions:read, profile:read). Which data a client can see is determined by the permission set defined when the token is created—not by the client itself.

Why this needs the data foundation that wealthAPI provides

An MCP server is only as robust as the data it can access. As a BaFin-regulated account information service provider connected to more than 3,500 banks and brokers, wealthAPI delivers the aggregated wealth view that enables the individual tools to provide reliable answers—whether the query is “How has my portfolio developed?” or a more complex, free-form question.

Context and next steps

The MCP server extends the existing wealthAPI API with an additional, AI-specific access path—it does not replace it. Fintechs, banks, and wealth managers already using wealthAPI for multibanking or investment data connectivity can use it to evaluate how AI-enabled access to the same data foundation can be integrated into their own products.

The technical details (requirements, token creation, configuration for Claude Desktop and Claude Code) are described in the API documentation.

Questions and answers

Is access secure?
The server runs locally and accesses data only via the four defined read scopes. There are no write functions and no way to access the system beyond the defined tools.

Which clients are supported?
All MCP clients that can integrate local servers, such as Claude Desktop and Claude Code. Purely web-based clients are not supported because there is no hosted server version.

How do you get an access token?
The token is currently created via the wealthAPI API; a UI for token creation is in preparation.

Is the code available to view?
Yes. The server is open source and available on GitHub.

What changes for existing API integrations?
Nothing. The MCP server uses the same data foundation as the REST API and is added as an additional access path.