How to Connect Your Brokerage to ChatGPT

A language model is not wired into your broker. It has no login to your account, no real-time market feed, and no memory of your positions unless something feeds them in. So a bare ChatGPT conversation can explain what a covered call is or reason through an allocation in the abstract, but it cannot answer “how is my portfolio doing” because it has never seen your portfolio. Worse, when pressed for specifics it can produce a confident, wrong figure, because it is pattern-completing rather than reading data.

Closing that gap means giving ChatGPT a grounded view of your real holdings. There are two honest ways to do that, plus a manual fallback. For the full explanation of the underlying limitation, see why ChatGPT can’t see your portfolio. The same wall exists with other models, so the pattern is the same when you connect a brokerage to Claude or any other AI assistant.

Under the hood, none of these approaches give ChatGPT your password. The mechanism is always the same three layers, and understanding them tells you what is safe and what is not.

• The account link (aggregator or broker API). A a secure aggregator or Plaid, or a broker’s own API such as Alpaca or Interactive Brokers, authenticates you with your broker through a token, not a shared password. It then exposes a read-only view of your positions. Most major brokers, including Robinhood, Charles Schwab, and Fidelity, can be linked this way for reading; only a subset allow placing trades.
• The bridge (MCP or a hosted assistant). That read-only feed has to reach the model. MCP, the Model Context Protocol, is the open standard a client uses to call external tools and data, so an MCP connector wraps the broker feed as tools ChatGPT can call. A connected assistant like Walnut does the same wrapping for you and hosts it, so there is no server for you to run.
• The model (ChatGPT). ChatGPT is the reasoning layer at the front. Once the bridge supplies your real positions, the model can answer questions about what you actually hold instead of guessing. On its own it is a smart explainer; with the link it becomes grounded in your data.

The one line worth remembering: the model never touches your broker directly. The aggregator or API holds the connection read-only, the bridge exposes it, and execution stays gated behind you. For the connector layer specifically, see the best MCP connectors for brokerages.

When you go to connect a brokerage to ChatGPT, these are the names that come up: the aggregators and broker APIs that hold the link, the brokers on the other end, and the one connected assistant that bundles it all. Here is what each is best for.

Two more names round out the picture. ChatGPT is best for the conversation itself: it is the reasoning front end once something feeds it your data. MCP (the Model Context Protocol) is best for the plumbing: it is the open standard that lets a client call your broker feed as a tool.

There is no single “Connect broker” button inside ChatGPT itself. Instead you use one of two bridges, or, if you just want a quick look, paste your holdings in. Here is what each does and what setting it up involves.

The MCP route gives you direct control and is the more technical path; a connected assistant does the wiring for you. For a deeper look at the connector option and what to look for in one, see MCP connectors for investing.

Whichever approach you choose, the shape of the process is the same. These steps keep it read-only and keep you in control of anything that touches your money.

• 1. Pick an approach. If you want direct control and do not mind running a server, set up an MCP connector. If you want it hands-off, use a connected assistant such as Walnut that links your broker for you. Pasting holdings into a message is the zero-setup fallback if you only need a one-time snapshot.
• 2. Connect read-only. Link your existing brokerage with read-only access, not trading access, to start. A connected assistant like Walnut connects through a regulated connection, read-only by default; an MCP connector should be authenticated with the narrowest scope the broker allows.
• 3. Verify the permissions. Before you ask anything, confirm what the tool can actually do: that it reads positions but cannot place orders without you, and that the access is scoped to viewing. If a tool wants trading permission up front, stop and understand why.
• 4. Ask about your holdings. Now the conversation is grounded. Ask how a position is doing, how concentrated you are, or how a holding compares to the S&P 500. With a connected assistant like Walnut you are talking through ChatGPT about your real, live positions rather than a stale snapshot.
• 5. Keep trade approval with you. Never let the model place an order unattended. The safe pattern is research and reasoning from the AI, execution from you. Walnut requires you to approve every trade, so nothing is bought or sold without an explicit yes from you.

To be upfront, since this is our site: Walnut is one of the connected-assistant options, not the only way and not a number one. Walnut is the AI investing assistant that talks to the broker you already have and places the trades you approve. It connects your existing brokerage and lets you talk through ChatGPT about your real connected holdings, with each position framed against the S&P 500, without you running any server yourself.

The point of the connected-assistant route is that it removes the setup: you link your broker read-only, and the chat is grounded in what you actually hold. Walnut is not a deep data terminal and not a budgeting app. It sits on top of your broker, leans on web and price data rather than a proprietary filings corpus, and because broker feeds rarely pass cost basis it frames returns as window returns rather than realized profit and loss, and says so. It is read-only by default, every trade needs your approval, and Walnut is not an investment adviser. Under the hood it is a hosted MCP connector for Claude, ChatGPT, and Codex (and any MCP client): you connect it by adding its server URL, https://walnutinvest.com/api/connector/mcp, as a custom connector, then signing in with a free Walnut account and approving read-only access, and it never places trades on its own. The full per-client setup lives on the Walnut MCP connector page. If you would rather keep direct control and do the wiring yourself, the MCP-connector route is the alternative.

Connecting your money to an AI tool is worth doing carefully. A few filters keep the whole thing trustworthy no matter which approach you use:

• Start read-only. Give the tool a view of your holdings, not the ability to trade. You can always keep it that way. Walnut is read-only by default.
• Prefer regulated aggregation. If a tool links your broker, prefer one that connects through a regulated aggregator rather than asking for your raw login. Walnut connects your brokerage securely.
• Keep execution human. The AI can research and frame trade-offs; the decision and the order should be yours. Approve every trade explicitly.
• Verify specific numbers. Even grounded, an AI can slip. Treat figures as a starting point and check anything you would act on against your broker.
• Watch the advice line. A trustworthy tool explains and frames without pretending to be your adviser. Be wary of anything promising guaranteed market-beating returns.

ChatGPT alone cannot see your brokerage, and it can state wrong figures with confidence, so the honest move is to bridge that gap deliberately. Set up an MCP connector if you want direct control and do not mind the technical setup, or use a connected assistant like Walnut if you want your broker linked for you and the chat grounded in your real holdings through ChatGPT. Either way, connect read-only, verify the permissions, ask your questions, and keep every trade behind your own approval. Walnut is not an investment adviser.

For more on the underlying limit and how to work around it, see how to use ChatGPT to analyze your portfolio and why ChatGPT can’t see your portfolio.