Brokerage to AI Assistant Integration: Providers Compared

Last updated October 2026

Short answer

If you are an investor connecting your own accounts, start with Walnut. It connects several brokers at once, gives each holding a keep, trim or exit read against the S&P 500, turns a thesis into a basket with target weights, and places the trades you approve at Public, plus eight other brokers including E*Trade, tastytrade and Webull, in its own assistant or alongside Claude and ChatGPT. The other end-user options each do part of that. The brokers' own AI covers one broker: in-app assistants such as Robinhood Agents and Schwab's Charley (announced September 30, 2026), and official connectors from Interactive Brokers (the AI drafts orders you submit), Robinhood, Webull, Public, tastytrade, Moomoo and TradeStation. ChatGPT Finances reads US accounts through Plaid and cannot trade (see Walnut vs ChatGPT Finances). If you are a developer or a brokerage choosing infrastructure, the field is SnapTrade (multi-broker read plus trading through an API) and Plaid (banking-first, read-only), or a single-broker API such as Alpaca or Interactive Brokers. Market-data platforms such as Alpha Vantage, Polygon, and Finnhub are not account integrations, because they connect no brokerage at all.

Invest with AI

“Brokerage integration with an AI assistant” is one phrase covering at least four different products. Some providers sell the plumbing. Some sell a finished platform. Some are a single broker with an API. And some sell market data that never touches your account and is mistaken for integration constantly. This guide compares the real options on the same fields (what the integration actually does, broker coverage, read-only versus trade-enabled, who it suits, and one honest limitation), explains how account linking works underneath, and is specific about where each provider, including Walnut, is the wrong choice.

What brokerage-to-AI integration actually means

An integration is the connection that lets an AI assistant reason about a real account instead of a made-up example. Strip away the marketing and every legitimate provider in this category implements roughly the same mechanism.

Authorization happens at the broker, not at the AI platform. You start the link inside the platform, get redirected to your brokerage's own authentication page, sign in there, and approve access. This is the OAuth-style pattern most people already know from “sign in with” buttons, and it is the reason your password never reaches the AI tool. The platform receives a revocable access token instead.
Scopes decide what the integration can do. A read scope covers positions, balances, and order history. A trade scope covers order submission and is requested separately, where the brokerage supports it at all. Keeping the two apart is what makes “read-only by default” a real technical state rather than a promise.
A normalized API sits between the broker and the assistant. Every brokerage models positions, symbols, and orders slightly differently. The provider normalizes them, so a platform writes one integration and reaches many brokerages instead of maintaining one per broker.
The AI reaches the account through tools, not through raw credentials. The platform exposes a defined set of functions the model can call (read the portfolio, price a security, prepare an order). The model never sees a token or a login. This is what an MCP connector standardizes; see what an MCP connector is for the protocol-level detail.
Revocation works from both ends. You can cut the connection off inside the platform, and most brokerages also let you revoke third-party access from your own account settings. An integration you cannot revoke at the broker is worth avoiding.

For a step-by-step walkthrough of doing this yourself, see how to connect your brokerage to an AI assistant, and for the security questions specifically, see whether it is safe to connect your brokerage to an AI.

The three layers of the integration market

Most confusion in this category comes from comparing providers that are not competing with each other. There are three layers, and you generally pick one, not all three.

Layer 1, end-user platforms. Walnut, which spans several brokers and can act on them with your approval; the brokers' own assistants and official connectors, which each cover one; and ChatGPT Finances, which reads US accounts and cannot trade. The integration is already built, and what you get is a working assistant pointed at an account. This is the layer almost everyone searching for this actually wants.
Layer 2, aggregation infrastructure. SnapTrade and Plaid. Their core products are APIs a developer integrates: the account-linking flow and normalized data. If you are a person with a brokerage account rather than a company building a product, this layer is mostly not what you are shopping for.
Layer 3, single-broker APIs. Alpaca and Interactive Brokers. No aggregation, because you hold the account at that broker and authenticate straight to it. Maximum control, maximum setup, and only one brokerage.

The integration providers compared

Each provider below is described on the same six fields: what it is, what the integration actually does, broker coverage, read-only or trade-enabled access, who it suits, and one honest limitation.

Walnut

An AI investing app that connects several of the brokerage accounts you already have and then acts on them with you. It reads each holding against the S&P 500 and gives a keep, trim or exit read, turns an investment thesis into a basket with target weights, places the trades you approve at Public and at eight other brokers including E*Trade, tastytrade and Webull, and can run recurring buys toward those target weights on a schedule you set. It works through its built-in assistant, and its MCP connector brings the same portfolio into Claude, ChatGPT and other AI clients.

Category: AI investing app on top of the brokers you already use (end user).
What the integration does: Handles the whole chain for you: secure account linking, a live read of your real positions at each connected broker, and a set of portfolio tools the AI can call. Your broker login stays with your broker and is never stored by Walnut. The connector for Claude and ChatGPT reads and analyzes; orders are placed in the Walnut app, at the broker, only after you approve them. Recurring buys spend cash already in that brokerage account, because Walnut never moves money.
Broker coverage: Several brokers at once. Trades you approve at Public, and at E*Trade, tastytrade, Webull (US and Canada), moomoo, Questrade, Wealthsimple, Trading 212 and Stake (Australia); Schwab, Fidelity, Vanguard and Interactive Brokers connect read-only for analysis.
Read-only or trade-enabled: Read-only by default; any order requires your explicit approval.
Who it is for: An investor who wants one AI that knows the portfolio they already hold across brokers and can act on it, with no code and no account transfer.
One honest limitation: It is a layer on top of a brokerage rather than a brokerage itself, so you need an existing account to connect, and trades route to one default trading account. Broker feeds rarely pass cost basis, so returns are framed as window returns rather than realized profit and loss, and it does not look inside the ETFs you hold.

Broker-native AI assistants and connectors (Robinhood, Schwab, Public, Webull, tastytrade, Moomoo, TradeStation and others)

AI shipped by brokerages themselves. In-app: Robinhood Cortex and Robinhood Agents (September 30, 2026), Schwab's Charley (announced September 30, 2026, rolling out in phases from October to eligible US retail clients on Schwab.com and Schwab Mobile), Public Alpha, Webull's AI signals. As official connectors for Claude, ChatGPT and other assistants, all launched in 2026: Robinhood's Agentic Trading MCP (May 27), Public's hosted MCP (Agentic Brokerage, March 31), Webull's MCP and native connectors (August, with order caps and a read-only mode), tastytrade's self-hosted server (September), Moomoo's hosted MCP, and TradeStation's MCP (January, needs a $10,000 balance and a paid Claude or ChatGPT plan).

Category: A broker's own AI, in its app or as an official connector.
What the integration does: In-app AI is not an integration at all: the AI and the account already live in the same platform. The official connectors are a real integration, run by the broker itself, so an outside assistant reaches the account with no aggregator in between. Either way, it sees only the accounts at that one broker.
Broker coverage: Only the broker's own accounts. Nothing you hold elsewhere is visible to it.
Read-only or trade-enabled: Varies by broker. Robinhood, Public and Webull let an agent place orders. Schwab's launch announcement lists answers on balances, positions and statements, guidance on next steps for trading, and select account tasks such as watchlists and alerts, not order placement; check tastytrade, Moomoo and TradeStation's documentation for their scope.
Who it is for: Someone who keeps everything at one broker that has its own connector or assistant.
One honest limitation: No cross-broker view, and you cannot bring the connector to a different broker. If your holdings are split across platforms, none of them sees the whole picture.

ChatGPT Finances

OpenAI's Finances feature in ChatGPT. US users link bank and investment accounts through Plaid, and ChatGPT can then read balances, holdings and daily changes and answer questions about them in the same chat.

Category: Account linking built into ChatGPT (end user, US only).
What the integration does: A real read integration, run by OpenAI on Plaid. You link accounts inside ChatGPT, sign in on the institution's own flow, and ChatGPT reads the data. Nothing is written back: it cannot place trades, move money or act on a target allocation.
Broker coverage: US institutions that Plaid supports; some share balances but not full holdings.
Read-only or trade-enabled: Read-only. It cannot place trades or move money.
Who it is for: A US user who wants a read-only, whole-money view (spending, bank balances and investments) inside ChatGPT.
One honest limitation: It reads and does not act: no trades, no target weights, no baskets, and nothing in Claude or outside the US. Compared in detail in Walnut vs ChatGPT Finances.

SnapTrade

An independent brokerage aggregation provider. Developers and fintech apps use its API to link a user's existing brokerage account, read positions, balances, and order history, and in many cases place orders back at that broker.

Category: Brokerage data and trading aggregator (developer infrastructure).
What the integration does: Provides the account-linking layer plus a normalized API, so one integration reaches many brokerages instead of one integration per broker. The user authenticates on the broker's own login screen and the platform receives a token, not a password.
Broker coverage: Broad multi-broker coverage across major US retail brokerages, with a smaller subset that supports order placement as well as read.
Read-only or trade-enabled: Both, depending on the broker: read-only for data-only brokerages, trade-enabled where the brokerage exposes a trading API.
Who it is for: Developers and platforms that want one integration covering many brokerages rather than negotiating each broker separately.
One honest limitation: Its core product is aimed at developers and platforms. It supplies account data and order routing, not portfolio analysis against a benchmark, target weights or a thesis-to-basket workflow, so those are left to whatever is built on top.

Plaid

The best-known account aggregation provider in consumer fintech. It is banking-first (checking, savings, transactions, identity, payments) and also offers investments coverage: holdings, securities, and investment transactions.

Category: Account aggregation provider (banking-first, developer infrastructure).
What the integration does: Provides a hosted link flow and an API that returns normalized account and holdings data. It is the plumbing behind a large share of budgeting, lending, and personal-finance platforms.
Broker coverage: Very broad institution coverage on the banking side; investment account coverage exists but is a smaller part of the platform than deposit accounts.
Read-only or trade-enabled: Read-only. It is a data aggregation provider, not an order-routing provider.
Who it is for: Products where investment data is one part of a wider financial picture that includes bank accounts, cash flow, and identity.
One honest limitation: No trade execution at all, and the investments product is not as deep on brokerage-specific detail as an aggregator built only for brokerage accounts.

Alpaca

A brokerage built API-first for developers and algorithmic traders, with market data, a paper-trading sandbox, and live execution. It also publishes an official open-source MCP server so AI agents can reach an Alpaca account directly.

Category: API-first broker (single broker, developer-oriented).
What the integration does: Because it is the broker, there is no aggregation layer in between: your API keys authenticate straight to your own Alpaca account. The MCP server exposes quotes, positions, and order placement as tools an AI client can call.
Broker coverage: One brokerage only, Alpaca itself. It cannot read or trade an account you hold anywhere else.
Read-only or trade-enabled: Read and trade-enabled, including a paper-trading mode for testing before anything touches real money.
Who it is for: Developers who are happy to open an Alpaca account and want an AI agent with full programmatic execution.
One honest limitation: You have to move to or open an account at Alpaca, and the MCP server is self-hosted, so there is setup, key handling, and maintenance you own.

Interactive Brokers

A large global brokerage with a long-standing, powerful API surface used by professional and advanced retail traders across a very wide range of instruments and markets.

Category: Full-service broker with its own API (developer-oriented).
What the integration does: An official AI connector: certified Claude and ChatGPT connectors by June 8, 2026, opened to any MCP client on July 28, 2026. The AI can read the account and draft orders, which land in an "AI Instructions" area for you to submit. Its long-standing API remains the route for fully programmatic trading.
Broker coverage: One brokerage, but an unusually broad one in instruments and geographies, which is why international and advanced investors gravitate to it.
Read-only or trade-enabled: Through the AI connector, read plus drafted orders that you submit; full trading through its API.
Who it is for: Advanced and international investors already at IBKR who want programmatic or agent-driven access to a deep instrument set.
One honest limitation: It covers IBKR accounts only, the AI connector does not execute on its own, and the full API has a real learning curve for anyone who wants programmatic trading. Through Walnut, IBKR is read-only.

Market-data MCP servers (Alpha Vantage, Polygon, Finnhub and similar)

Data providers and the MCP servers built around them. They give an AI assistant quotes, fundamentals, news, and historical prices through an API key.

Category: Market data APIs, not account integrations.
What the integration does: There is no account linking anywhere in this category. The assistant gets market data, not your data, and confusing the two is the single most common misunderstanding when people search for a brokerage integration provider.
Broker coverage: Securities and market data coverage, not brokerage account coverage. Zero brokers are connected.
Read-only or trade-enabled: Read-only, and read-only on public market data. No positions, no balances, no orders.
Who it is for: Adding live market context to an AI assistant, often alongside a separate account integration.
One honest limitation: The assistant still cannot see what you own, so every answer about your portfolio is a guess unless you pair it with a real account integration.

At a glance

Provider	Category	Broker coverage	Read-only or trade-enabled
Walnut	AI investing app on top of the brokers you already use (end user)	Several brokers at once. Trades you approve at Public, and at E*Trade, tastytrade, Webull (US and Canada), moomoo, Questrade, Wealthsimple, Trading 212 and Stake (Australia); Schwab, Fidelity, Vanguard and Interactive Brokers connect read-only for analysis	Read-only by default; any order requires your explicit approval
Broker-native AI assistants and connectors (Robinhood, Schwab, Public, Webull, tastytrade, Moomoo, TradeStation and others)	A broker's own AI, in its app or as an official connector	Only the broker's own accounts. Nothing you hold elsewhere is visible to it	Varies by broker. Robinhood, Public and Webull let an agent place orders. Schwab's launch announcement lists answers on balances, positions and statements, guidance on next steps for trading, and select account tasks such as watchlists and alerts, not order placement; check tastytrade, Moomoo and TradeStation's documentation for their scope
ChatGPT Finances	Account linking built into ChatGPT (end user, US only)	US institutions that Plaid supports; some share balances but not full holdings	Read-only. It cannot place trades or move money
SnapTrade	Brokerage data and trading aggregator (developer infrastructure)	Broad multi-broker coverage across major US retail brokerages, with a smaller subset that supports order placement as well as read	Both, depending on the broker: read-only for data-only brokerages, trade-enabled where the brokerage exposes a trading API
Plaid	Account aggregation provider (banking-first, developer infrastructure)	Very broad institution coverage on the banking side; investment account coverage exists but is a smaller part of the platform than deposit accounts	Read-only. It is a data aggregation provider, not an order-routing provider
Alpaca	API-first broker (single broker, developer-oriented)	One brokerage only, Alpaca itself. It cannot read or trade an account you hold anywhere else	Read and trade-enabled, including a paper-trading mode for testing before anything touches real money
Interactive Brokers	Full-service broker with its own API (developer-oriented)	One brokerage, but an unusually broad one in instruments and geographies, which is why international and advanced investors gravitate to it	Through the AI connector, read plus drafted orders that you submit; full trading through its API
Market-data MCP servers (Alpha Vantage, Polygon, Finnhub and similar)	Market data APIs, not account integrations	Securities and market data coverage, not brokerage account coverage. Zero brokers are connected	Read-only, and read-only on public market data. No positions, no balances, no orders

Coverage: which brokers can actually be integrated

Coverage is the first question worth asking any provider, and it has two halves that people routinely collapse into one.

Read coverage is wide. Through a regulated aggregation provider, most major US retail brokerages link for read access, enough for the majority of what an AI assistant is useful for: reading what you hold, measuring each position against a benchmark, spotting concentration, and answering questions about your real portfolio rather than a hypothetical one.
Trade-enabled coverage is narrower, and it is the broker's decision. A brokerage has to expose a trading API before any integration can submit an order. Several large, well-known brokerages deliberately do not, so they sync holdings flawlessly and then reject every order, which comes down to the broker, not a failing of the provider.
The trade-enabled set differs by provider and moves over time. Aggregators add and drop partners, and the list is not the same across SnapTrade, a single-broker API, and any platform built on top. This is exactly why this page does not publish a broker-by-broker trade table: it would be wrong within months. Check your own brokerage against the provider you intend to use.
Being read-only is often fine. If your goal is analysis, tracking, and a second opinion, read access covers all of it and you place orders yourself in your broker's app. Only assume you need trade-enabled coverage if you specifically want to act from inside the assistant.

For what each broker ships in its own app rather than through an integration, see which brokers have an AI assistant.

Which provider should you pick?

For an investor connecting their own accounts, Walnut is the first pick, because it is the one end-user platform here that spans several brokers and also acts: trades you approve at Public and eight other brokers including E*Trade, tastytrade and Webull, baskets with target weights, and recurring buys toward them. Developers and algorithmic traders are buying a different layer, so their picks are listed separately. Within each use case, the stronger fit is listed first.

Best integration for an investor connecting their own accounts

If you do not want to write code, the question is which platform already did the integration work for you and put an AI assistant on top of it, and what that assistant can do once it sees your account.

Walnut. The one that reads and also acts: several brokers at once, a keep, trim or exit read on each holding against the S&P 500, a thesis turned into a basket with target weights, and trades you approve at Public and eight other brokers including E*Trade, tastytrade and Webull, including recurring buys toward those targets. Works in its own assistant, Claude and ChatGPT.
Broker-native AI assistants and connectors (Robinhood, Schwab, Public, Webull, tastytrade, Moomoo, TradeStation and others). If everything you own sits at one broker with its own assistant or official connector (Robinhood, Schwab's Charley, Public, Webull and others), that is the least effort of all for that one account.
ChatGPT Finances. A read-only US view of bank and investment accounts inside ChatGPT. It cannot trade or work toward target weights.

Best integration provider for a developer building on many brokers

If you are building a platform and want one integration that reaches many brokerages, you want an aggregator with an API, not an end-user product.

SnapTrade. Purpose-built brokerage aggregation: one API for multi-broker read plus order placement where the brokerage supports it.
Plaid. The broadest institution coverage overall, and the right pick when investment data sits next to bank accounts and cash flow. Read-only, no execution.

Best integration for a developer who wants full agent execution

If you want an AI agent that can actually place orders programmatically and you are willing to hold the account at a specific broker, go straight to a broker API.

Alpaca. API-first broker with an official open-source MCP server, live and paper trading, so an agent can be tested before it touches real money.
Interactive Brokers. The deepest instrument and geography coverage of any single-broker API, if you can absorb the complexity. Its official AI connector drafts orders; full automation goes through the API.

Best for adding market data to an AI assistant

Worth separating out, because this is not account integration and buying it expecting portfolio awareness is the classic mistake.

Market-data MCP servers (Alpha Vantage, Polygon, Finnhub and similar). Quotes, fundamentals, and news through an API key. Pair it with a real account integration if you want the assistant to reason about what you actually hold.

By kind of user

The same field maps onto who you are as cleanly as it maps onto use case:

The DIY investor with accounts at more than one broker. You want a platform, not an API, and you want cross-broker coverage, so a broker's own assistant or connector will always show you a partial picture. An end-user platform that links each account read-only and reasons across all of them is the fit. Walnut and the wider field are compared in the best apps to connect a brokerage to an AI assistant roundup.
The developer building a product. You are buying infrastructure, so the questions are coverage, API quality, normalization, and whether trading is supported. SnapTrade for multi-broker brokerage data and orders, Plaid when banking data dominates the picture, Alpaca when you would rather be the broker relationship yourself.
The person who just wants to chat with their portfolio. You do not care about any of the above. You want to link an account in a few clicks and ask questions in plain English. Pick an end-user platform, confirm it is read-only by default, and confirm your brokerage is covered before you sign up.
The algorithmic trader. You want deterministic execution and a sandbox to test in, which points at a single-broker API rather than an aggregator. Alpaca for the cleanest developer experience, Interactive Brokers for instrument and geographic breadth.

How we evaluated these providers

We limited the field to things that actually bear on connecting a brokerage account to an AI assistant, which is why market-data platforms are included as a category to rule out rather than as a competitor. Within that set we weighed five things:

Coverage: how many brokerages the integration reaches, and how much of that is trade-enabled rather than read-only.
Credential model: whether authentication happens on the broker's own page and the platform holds a revocable token rather than a password.
Scope granularity: whether read and trade are genuinely separable, and whether orders sit behind human approval.
Who it is actually for: whether the thing is an API a company integrates or a product a person uses, since the two get compared as if interchangeable.
Honesty about limits: we marked down anything implying an integration can do more than the underlying broker permits.

For investors we ranked Walnut first, because it is the only option here that covers several brokers and acts on them with your approval; the developer layers are ranked separately because they do not compete with it. Features, coverage, and pricing change; treat this as a starting point and verify current details on each provider's own site.

Where Walnut fits

To be upfront, since this is our site: Walnut is an end-user platform, the first layer above. It is a free AI investing app that connects the brokerages you already use, several at once, and does four jobs on them: a keep, trim or exit read on each holding against the S&P 500, a thesis turned into a basket with target weights, trades you approve, and recurring buys toward those targets that spend cash already in your brokerage account (Walnut never moves money). It works through its built-in assistant or through Claude, ChatGPT, and other MCP clients, where the connector reads and analyzes and orders stay in the Walnut app. ChatGPT Finances differs here: it reads US accounts and cannot act on them. The integration work is already done, so there is no API to code against and no server to run. Your broker login stays with your broker, the connection is read-only by default, and where the connected brokerage is trade-enabled, Walnut prepares orders and routes them back to that broker for you to approve (today Public, plus E*Trade, tastytrade, Webull (US and Canada), moomoo, Questrade, Wealthsimple, Trading 212 and Stake (Australia)). It also detects when a connected account is read-only and tells you, rather than letting orders fail one at a time. Because broker feeds rarely pass cost basis, returns are framed as window returns rather than realized profit and loss, and Walnut says so on the page. Walnut is not a registered investment adviser and nothing it produces is a recommendation to buy or sell.

Where Walnut is the wrong choice

Just as importantly, here is when a different provider fits the job better:

You are building a product and need the raw integration. Walnut is a finished platform, not an API you embed. SnapTrade for brokerage read and trading, Plaid when banking data is the centre of gravity.
You want an agent that executes autonomously. Walnut keeps every order behind your approval by design. If you want execution without a human in the loop, a single-broker option like Alpaca's API or Robinhood's agents is the right layer.
You do not want to connect a brokerage account at all. Walnut sits on top of a real account, so it needs one. A market-data MCP server gives an assistant prices and fundamentals with no account linking whatsoever.
Your brokerage is not covered, or is read-only and you specifically want to trade from the assistant. Check coverage before signing up. If your broker exposes no trading API, no integration provider can change that.
You hold everything at one broker that has its own connector or assistant. Robinhood, IBKR, Webull, Public, tastytrade, Moomoo and TradeStation now offer official connectors for their own accounts, and Schwab is rolling out its Charley assistant inside Schwab.com and Schwab Mobile, so that is a fine choice. Walnut connects Schwab read-only, for analysis rather than trading.
You want managed money rather than an assistant. A robo-advisor builds and runs a portfolio for you, a different category entirely, and no amount of integration turns an assistant into one.

From a connected account you can dig into a specific stock, an ETF you hold, or a theme you want exposure to.

Get a recommendation for your situation

Walnut is the AI that knows your portfolio: ask anything in plain English, research any fund, and get an honest second opinion. On the broker you already use, read-only, and you approve every trade. Walnut is not a registered investment adviser.

Invest with AI

FAQ

Which providers offer brokerage integration with AI assistants?

For an investor, start with Walnut: it connects several brokers, reads each holding against the S&P 500, and places the trades you approve at Public, plus eight other brokers including E*Trade, tastytrade and Webull, in its own assistant or alongside Claude and ChatGPT. Next come the brokers' own AI (Robinhood, Schwab's Charley, Public, and official connectors from IBKR, Webull, tastytrade, Moomoo and TradeStation), each for its own accounts, and ChatGPT Finances for a read-only US view. Developers build on SnapTrade or Plaid, or on Alpaca's and IBKR's APIs.

Which company has the best brokerage-to-AI assistant integration?

For an investor with accounts at one or more brokers, Walnut. It is the integration that reads and also acts: a keep, trim or exit read per holding against the S&P 500, a thesis turned into a basket with target weights, and trades you approve at Public, plus eight other brokers including E*Trade, tastytrade and Webull (US and Canada). If everything sits at one broker with its own assistant, such as Robinhood or Schwab's Charley, that is the least effort. Developers building a product want SnapTrade or Plaid instead.

What is a brokerage-to-AI integration?

It is the connection that lets an AI assistant see and act on a real brokerage account instead of guessing. Technically it is an account-linking flow plus an API: you authorize access on the broker's own login screen, the provider receives a token, and the AI platform reads positions and balances through that token. Without an integration, Claude cannot see any of your holdings. ChatGPT can for US users who link accounts through its Finances feature (built on Plaid), but it cannot make trades.

How does the integration work technically?

You start the link inside the AI platform, get redirected to your broker's own authentication page, sign in there, and approve a scoped grant. The provider stores an access token, never your password. The platform then calls the provider's API for positions, balances, and order history, and exposes those as tools the AI model can call. Trade scopes, where the broker supports them, are separate from read scopes and require you to approve each order.

Do my broker credentials ever reach the AI tool?

With a properly built integration, no. You authenticate on the broker's own domain and the platform receives a revocable token instead of a password, and that token model is the main security property to look for. If a tool asks you to type your brokerage username and password into its own form, it is not using this model, and you should treat that as a red flag. You can also revoke the connection at the broker or in the platform at any time.

What is the difference between read-only and trade-enabled integration?

Read-only means the assistant can see holdings, balances, and history and analyze them, but cannot move anything. Trade-enabled means the integration can also submit orders. Most integrations are read-only, and most that support trading keep it behind explicit per-order approval. Plaid is read-only by definition. SnapTrade supports both depending on the brokerage. Interactive Brokers' AI connector sits in between: it drafts orders that you submit. Walnut is read-only by default and requires your approval for every order.

Which brokers can actually be integrated?

Through a regulated aggregator, most major US retail brokerages connect for read access, which covers analysis, tracking, and benchmarking. Trade-enabled coverage is a narrower subset, because a brokerage has to expose a trading API for orders to be possible at all, and several large brokerages deliberately do not. The trade-enabled set also differs by provider and changes as aggregators add partners, so verify your specific broker rather than trusting any published list, including this one.

Why are some brokers read-only?

Because reading positions and placing orders are separate capabilities, and a brokerage can support one without the other. A brokerage that has no external trading API can still be aggregated for data, so the integration syncs holdings perfectly and then rejects every order. The cause lies with the broker, not a limitation of the integration provider. Good platforms detect it up front and tell you, rather than letting orders fail one by one.

Is SnapTrade or Plaid better for investment account integration?

They solve different problems. SnapTrade is built specifically for brokerage accounts and is the stronger choice when you need position-level detail and the ability to place orders. Plaid has the broadest institution coverage in consumer fintech and is the stronger choice when investment data is one part of a wider financial picture that includes bank accounts, transactions, and identity. Plaid is read-only, so it cannot route trades at all.

Are Alpha Vantage, Polygon, and Finnhub brokerage integrations?

No. Those are market-data providers, and the MCP servers built around them give an AI assistant quotes, fundamentals, news, and historical prices. There is no account linking anywhere in that category, so the assistant still cannot see a single share you own. They are useful alongside a real account integration, not instead of one, and confusing the two is the most common mistake in this category.

Do broker-native AI assistants count as brokerage-to-AI integration?

In-app assistants such as Robinhood Cortex, Schwab's Charley and Public Alpha do not, because nothing is being integrated. The official connectors brokers launched in 2026 do: Robinhood, Webull, Public, IBKR, tastytrade, Moomoo and TradeStation each let Claude, ChatGPT or another assistant reach your account at that broker. Either way, they cover one broker only, so they are the lowest-effort option if you hold everything there and a partial picture if you do not.

Do I need to be a developer to use one of these integration providers?

Only for part of the field. SnapTrade, Plaid, Alpaca, and the Interactive Brokers API are developer infrastructure: you write code against them. An end-user platform like Walnut has already done that integration work, so linking a brokerage is a few clicks and a login on your broker's page. Broker-native assistants need no setup at all. Match the layer to whether you are building a product or using one.

Can ChatGPT or Claude integrate with my brokerage directly?

Partly, in 2026. In the US, ChatGPT's Finances feature can link accounts through Plaid and show balances and holdings, but it cannot make trades, holdings may be missing at some institutions, and it is US-only. Several brokers (IBKR, Robinhood, Webull, Public and others) now have official connectors for Claude and ChatGPT, each for its own accounts. Walnut exposes your portfolio across many brokers to Claude, ChatGPT, or other MCP clients, and Alpaca publishes an official MCP server for its own accounts.

Can an AI assistant place trades through the integration?

Only where the brokerage is trade-enabled and only if the platform requests trade scopes, and in practice the responsible platforms keep every order behind explicit human approval. Alpaca can execute programmatically, as can the official Robinhood, Public and Webull connectors on their own accounts; Interactive Brokers' AI connector drafts orders for you to submit. Walnut prepares orders and routes them to your own broker for you to approve, and is read-only by default. Nothing here is autonomous trading, and none of it is a recommendation to trade.

What should I look for when choosing an integration provider?

Five things. Broker coverage, meaning whether your specific brokerage is supported and whether it is read-only or trade-enabled. The credential model, meaning whether you authenticate on the broker's own page rather than handing over a password. Scope granularity, meaning whether read and trade are separable and whether orders need approval. Data depth, meaning positions and history rather than a balance total. And revocability, meaning how easily you can cut the connection off at either end.

Walnut is informational and is not an investment adviser. Provider features, broker coverage, API capabilities, and pricing change; verify current details on each provider's site before deciding. Nothing on this page is a recommendation to buy, sell, or hold any security or to use any particular product.

Start here instead

This page covers one slice of the subject. For the full comparison, read How to connect your brokerage to an AI assistant, which is the main guide to connecting a brokerage to an AI assistant and the best place to start if you are still deciding.

ETFs and stocks in this guide

Stocks: IBKR