How to Connect Your Brokerage to Claude

Start from the honest constraint, because it changes what “connecting” means. Claude is a general-purpose assistant. It is excellent at explaining a concept, weighing a trade-off, or structuring a decision, but it has no built-in link to your brokerage and no feed of real-time quotes. Ask it about your portfolio in a plain chat and it can only reason from what you paste in or what it can search, not from your actual positions.

So connecting a brokerage to Claude is never a toggle inside Claude. It always means adding a layer that reads your accounts and makes that data available to the model. The layer is where the connection, the permissions, and the safety live, and it is the part worth choosing carefully.

There are two real ways to ground Claude in your holdings, plus a manual fallback that needs no connection at all. Each is described on the same two fields, what it does and how you set it up:

An MCP connector keeps you in Claude and adds a tool that reads your broker. A connected assistant handles the connection for you and routes your questions through Claude. Pasting in your holdings is the simplest path and the least automated. The rest of this guide walks the steps that apply to whichever you choose.

Decide how much wiring you want to do before you connect anything.

• MCP connector. If you want to keep chatting in Claude directly and are comfortable authorizing a tool, an MCP connector exposes your broker or portfolio data to Claude through the Model Context Protocol. You keep control of the chat; the connector is just the bridge. See our overview of MCP connectors for investing for how these work.
• Connected assistant that routes through Claude. If you would rather not wire anything up, a product like Walnut links your broker for you and lets you ask about your real holdings through Claude in its own chat. It handles the connection and the grounding; you just talk.
• Paste your holdings in. If you want zero connection, copy your positions into a Claude chat and let it reason over the text. Simple, but Claude only knows what you pasted, and you re-paste whenever things change.

For the general version of this decision across any model, see how to connect a brokerage to an AI assistant.

Whichever approach you chose, connect with the least access that does the job, which for asking questions is read-only. Reading your holdings is enough for Claude to reason over your portfolio; it does not need the ability to move money to answer “how is this position doing” or “what themes am I heavy in.”

Prefer a connection that runs through a regulated data aggregator rather than one that asks for your raw broker password. A connected assistant like Walnut connects your existing broker through SnapTrade and reads your holdings read-only by default, so your credentials go to the aggregator, not to the assistant or to Claude. If you are setting up an MCP connector yourself, choose one with the same read-only-first posture.

Before you rely on anything, confirm what you actually granted. This is the step people skip, and it is the one that matters most.

• Scope. Check that access is read-only unless you deliberately intend to enable trading. If a connector asks for trade permission to answer questions, that is a mismatch worth pausing on.
• What is shared. Confirm the layer reads holdings and not, say, your login credentials directly. A regulated aggregator keeps your password out of the assistant and out of Claude.
• Revocability. Make sure you can disconnect the account and revoke access from the broker or aggregator side, not just from the app.

If any of those are unclear, stop and read the tool’s security and permissions documentation before you keep going. The whole point of the layer is that you can trust what it can and cannot do.

Once the connection is live, you can finally do the thing you came for: ask Claude about your real portfolio in plain language. With an MCP connector, you ask in a normal Claude chat and it pulls your holdings as a tool. With a connected assistant, you ask in its chat and it routes the question through Claude with your positions already in context.

Good questions are the ones a general chat could never answer without your data: which positions are lagging, where you are concentrated, how a holding has moved over a window, or what a theme you are considering would add to what you already own. Walnut frames each position against the S&P 500 so the answers are relative, not just absolute, and pairs the chat with web search for context. As always, verify any specific figure before acting on it.

Reasoning over your holdings and executing on them are two different things, and the safe setup keeps them separate. Let Claude explain, compare, and frame trade-offs, but keep the decision and the order in your hands. Any trade should happen at your broker, behind your own explicit approval, never silently on the model’s say-so.

Walnut is built around exactly this line: it is read-only by default, and every trade requires your approval before anything is placed at your broker. Claude describes and reasons; you decide. Walnut is informational and is not an investment adviser, so treat its output as research to check, not instructions to follow.

To be upfront, since this is our site: Walnut is one of the connected-assistant approaches above, not the only way and not the number one choice for everyone. If you want to keep chatting in Claude directly and are happy to wire up a tool, a dedicated MCP connector may suit you better. Walnut is the low-setup path.

Walnut is an AI investing assistant you chat with on the broker you already own. It connects your existing brokerage through SnapTrade, reads your holdings read-only by default, and lets you talk through Claude, ChatGPT, or a built-in assistant about what you actually hold, with each position framed against the S&P 500. It is not a deep data terminal and not a budgeting app, and because broker feeds rarely pass cost basis it frames returns as window returns rather than realized profit and loss, and says so. Every trade needs your approval, and Walnut is not an investment adviser.

For more on using Claude specifically as your research model, see Claude for investing.

Claude cannot connect to your brokerage by itself, so “connecting Claude to your broker” really means choosing the layer that sits between them. An MCP connector exposes your data to Claude while you keep chatting directly; a connected assistant like Walnut links your broker for you and routes the conversation through Claude; and pasting your holdings in is the manual fallback. Whichever you pick, connect read-only, verify the permissions, then ask Claude about your real holdings, and keep every trade behind your own approval. Walnut is not an investment adviser.