Best MCP Connectors for Brokerages in 2026
Last updated July 2026
Short answer
Only a few MCP servers can actually read and trade a real brokerage. Alpaca's official open-source server does data, paper, and live trading on Alpaca; community connectors cover Interactive Brokers, TradeStation, and Tradier; and Walnut is the no-code option that securely links most US brokers, reaching Schwab and Fidelity, and works from Claude or ChatGPT read-only by default. Market-data MCPs (Alpha Vantage, Polygon, Finnhub, Financial Modeling Prep, Yahoo Finance) only serve prices and fundamentals and cannot touch an account. There is no single best one: pick by your broker and whether you want execution.
Since Anthropic introduced the Model Context Protocol, a wave of finance MCP servers has shipped, and they are not interchangeable. Some connect to your actual brokerage so an assistant can read positions and place trades; most only serve market data. Confusing the two is the most common mistake, because a data MCP can never act on your portfolio no matter how good its prices are. This guide separates execution connectors from data servers, describes eleven of them on the same fields, ranks them by what you want to do, and is honest about where Walnut is and is not the right fit.
How an MCP connector bridges a broker to an AI client
An MCP connector for a brokerage is a server that exposes your account to an AI assistant as MCP tools (see what an MCP connector is for the plain-language version). The Model Context Protocol is an open standard, introduced by Anthropic in late 2024, for connecting assistants like Claude or ChatGPT to external data and tools through a uniform interface. A brokerage MCP server implements that interface against a broker's API: it advertises tools such as “get positions”, “get quotes”, and, on execution servers, “place order”. The assistant calls those tools during a conversation, so it can reason about your real holdings instead of a hypothetical example. The bridge has three moving parts: the AI client (Claude Desktop, ChatGPT, or Cursor), the MCP server that speaks the protocol, and the broker connection the server authenticates against.
Under the hood, a brokerage connector reaches your account one of two ways: directly through the broker's own API (Alpaca, Tradier, TradeStation, Interactive Brokers), or through a regulated account-aggregation layer such as SnapTrade or Plaid that fronts many institutions behind one connection. The aggregator route is how a single connector can reach brokers like Schwab and Fidelity without a bespoke integration for each. This is also the practical difference between an MCP connector and a raw broker API or plugin, covered in MCP connector vs API vs plugin.
The key split is what those tools can do. A read-and-trade server (Alpaca, IBKR connectors, Tradier, TradeStation, Walnut) can both report your account and submit orders. A market-data server (Alpha Vantage, Polygon, Finnhub, Financial Modeling Prep, Yahoo Finance) only returns prices and fundamentals and has no concept of your account at all. The rest of this guide treats those as two different categories, because comparing them on one axis is misleading.
What to look for in a brokerage MCP connector
- Broker coverage. Whether it works with your broker. Single-broker servers (Alpaca, Tradier) bind to one account; aggregator-based connectors (like Walnut) reach many.
- Read vs execution. Whether it only reads holdings or can also place trades, and what approval controls gate orders before they reach the broker.
- Auth and security model. Whether it uses scoped API tokens or a regulated aggregator, defaults to read-only, and never stores your broker password.
- Paper trading. A sandbox to test agent behavior before risking real money. Alpaca and Tradier offer one; most data servers are irrelevant here.
- Setup effort. A self-hosted open-source server (needs code and a host to secure) versus a hosted, no-code connector.
- Official vs community. Whether the broker maintains it (Alpaca) or it is an unofficial community wrapper (IBKR, Tradier, Yahoo Finance), which affects maintenance and trust.
Which MCP servers can actually trade a real account?
The brokerage-execution servers below connect to a real account. Each is described on the same fields: what it is, whether it reads or trades, which brokerage it covers, its auth and security model, who it suits, and one honest limitation. This is a fast-moving space, so newer and community names (for example Connect Trade) surface in AI answers too; verify their broker support, maintenance, and security before pointing any of them at a live account.
Alpaca MCP server (official)
Alpaca's official, open-source MCP server. It exposes Alpaca's trading API to an AI agent as MCP tools: real-time and historical market data, account and position reads, order placement, and a full paper-trading sandbox.
- Read or trade: Read + trade + paper trading.
- Brokerage: Alpaca brokerage accounts only.
- Auth and security: You supply your own Alpaca API key and secret as environment variables and self-host the server; credentials live in your config, not with a third party.
- Best for: Developers building agent-driven, automated workflows who already use (or will open) an Alpaca account.
- One honest limitation: Single-broker: it only touches Alpaca accounts, and you run and secure the server yourself.
Interactive Brokers (community MCP connectors)
IBKR has no official MCP server, but the community has built open-source connectors that wrap the IBKR API (the Client Portal Web API or the TWS/Gateway). They expose account reads, positions, market data, and order placement to an agent.
- Read or trade: Read + trade (paper and live, depending on the connector).
- Brokerage: Interactive Brokers accounts.
- Auth and security: You authenticate through IBKR's own gateway or Client Portal session, which keeps login at IBKR; the MCP layer talks to that local session.
- Best for: Advanced and international IBKR users comfortable running their own gateway and vetting community code.
- One honest limitation: Community-built and unofficial, so quality, maintenance, and security vary by project; you must run the IBKR gateway alongside it.
Walnut connector
Walnut's portfolio-aware MCP connector. You securely connect a real broker you already have once, then read your live portfolio and use thematic-portfolio tools from Claude Desktop, ChatGPT, or Cursor. Walnut offers both a hosted connector and a small downloadable local MCP shim, so you can use whichever your AI client supports, with no server code to write.
- Read or trade: Read by default; trades only with your explicit approval (and only where the broker supports execution).
- Brokerage: Most major US brokers for reading (Schwab, Fidelity, and more), with trade execution on a subset (for example Alpaca, Public, Schwab, Tradier, Webull); some, like Robinhood, connect read-only.
- Auth and security: Your broker login and credentials stay at the broker; the connector authenticates with a per-user Walnut key and access is read-only unless you enable trading.
- Best for: People who want broad US-broker coverage and a no-code setup from Claude Desktop, ChatGPT, or Cursor rather than self-hosting a broker server.
- One honest limitation: It sits on top of your broker rather than being one, so you need an existing account, and broker feeds rarely pass cost basis (returns are framed as window returns, not realized profit and loss). Walnut is not an investment adviser.
Tradier (community MCP connectors)
Tradier offers a developer-friendly REST brokerage API, and community MCP servers wrap it so an agent can read accounts, pull quotes and options chains, and place equity and options orders.
- Read or trade: Read + trade (sandbox and live).
- Brokerage: Tradier brokerage accounts.
- Auth and security: You generate a Tradier access token and pass it to the self-hosted server; the token scopes what the agent can do.
- Best for: Developers who want a simple REST broker with strong options support behind an agent.
- One honest limitation: Community-built rather than official, single-broker, and you self-host and secure it.
TradeStation (community MCP connectors)
TradeStation publishes a full brokerage REST API covering equities, options, and futures, and community MCP servers wrap it so an agent can read accounts, pull quotes, and place orders.
- Read or trade: Read + trade (simulated and live).
- Brokerage: TradeStation brokerage accounts.
- Auth and security: You authenticate with TradeStation OAuth credentials scoped to your account and pass them to the self-hosted server; the scope limits what the agent can do.
- Best for: Active traders on TradeStation who want equities, options, and futures behind an agent.
- One honest limitation: Community-built rather than official, single-broker, and you run and secure the server yourself.
CCXT / crypto-exchange MCP servers
A family of MCP servers built on the CCXT library (and exchange-specific ones) that connect crypto exchanges like Coinbase, Kraken, and Binance to an agent for balances, market data, and order placement.
- Read or trade: Read + trade (varies by exchange and key permissions).
- Brokerage: Crypto exchanges, not equity brokers.
- Auth and security: You provide exchange API keys with scopes you control (read-only or trade); keys live in your self-hosted config.
- Best for: Builders automating crypto trading or tracking across exchanges with one MCP interface.
- One honest limitation: Crypto exchanges only, not stock brokerages, and a misconfigured key scope can grant more access than intended.
Which MCP servers are market-data only?
These do not connect to a brokerage. They feed an assistant prices, fundamentals, and news so its analysis is grounded in real data rather than stale training data. They cannot read or act on your account. Many people pair one of these for research with a brokerage-execution server for the portfolio itself. Same field block as above.
Alpha Vantage MCP server
A market-data MCP server fronting the Alpha Vantage API. It gives an agent broad equity prices, fundamentals, technical indicators, forex, and crypto data so its analysis runs on real numbers instead of stale training data.
- Read or trade: Data only (no account, cannot trade).
- Brokerage: None (market data).
- Auth and security: A free or paid Alpha Vantage API key in your config; no brokerage access at all.
- Best for: Grounding an assistant's research in real prices and fundamentals without touching an account.
- One honest limitation: Pure data: it cannot see or act on any portfolio, and free-tier request limits are tight.
Polygon.io MCP server
A market-data MCP server for Polygon.io, which provides low-latency real-time and historical market data across stocks, options, forex, and crypto, aimed at data-intensive applications.
- Read or trade: Data only (no account, cannot trade).
- Brokerage: None (market data).
- Auth and security: A Polygon API key (paid tiers for real-time depth) in your config; no brokerage access.
- Best for: Low-latency, high-volume market data feeding an analytical agent.
- One honest limitation: Data only and the useful real-time tiers are a paid subscription.
Financial Modeling Prep MCP server
A market-data MCP server fronting Financial Modeling Prep, which leans toward fundamentals: financial statements, ratios, valuations, earnings, and company profiles alongside prices.
- Read or trade: Data only (no account, cannot trade).
- Brokerage: None (market data).
- Auth and security: An FMP API key in your config; no brokerage access.
- Best for: Fundamentals-heavy research (statements, ratios, valuation) for an agent.
- One honest limitation: Data only, and deeper fundamentals history sits behind paid tiers.
Finnhub MCP server
A market-data MCP server fronting the Finnhub API, which serves real-time quotes, company fundamentals, news, earnings, and alternative datasets so an agent reasons over live numbers instead of stale training data.
- Read or trade: Data only (no account, cannot trade).
- Brokerage: None (market data).
- Auth and security: A free or paid Finnhub API key in your config; no brokerage access at all.
- Best for: Real-time quotes, news, and fundamentals for an assistant, with a usable free tier.
- One honest limitation: Data only, and the higher-frequency and premium datasets sit behind paid tiers.
Yahoo Finance MCP servers (community)
Community MCP servers that wrap public Yahoo Finance endpoints for quotes, historical charts, and basic fundamentals, popular because they need no API key and are quick to stand up.
- Read or trade: Data only (no account, cannot trade).
- Brokerage: None (market data).
- Auth and security: Usually no key required; they read public, unofficial Yahoo endpoints.
- Best for: A free, zero-setup data feed for prototypes and personal projects.
- One honest limitation: Unofficial and unsupported: the public endpoints can change or rate-limit without notice, so it is not production-grade.
MCP brokerage connectors at a glance
| Connector | Read / trade | Brokerage | Auth model | Who it's for |
|---|---|---|---|---|
| Alpaca MCP server (official) | Read / trade | Alpaca brokerage accounts only | Self-host, broker API key | Developers building agent-driven, automated workflows who already use (or will open) an Alpaca account |
| Interactive Brokers (community MCP connectors) | Read / trade | Interactive Brokers accounts | Self-host, broker API key | Advanced and international IBKR users comfortable running their own gateway and vetting community code |
| Walnut connector | Read / trade | Most major US brokers for reading (Schwab, Fidelity, and more), with trade execution on a subset (for example Alpaca, Public, Schwab, Tradier, Webull); some, like Robinhood, connect read-only | Secure broker link, read-only default | People who want broad US-broker coverage and a no-code setup from Claude Desktop, ChatGPT, or Cursor rather than self-hosting a broker server |
| Tradier (community MCP connectors) | Read / trade | Tradier brokerage accounts | Self-host, broker API key | Developers who want a simple REST broker with strong options support behind an agent |
| TradeStation (community MCP connectors) | Read / trade | TradeStation brokerage accounts | Self-host, broker API key | Active traders on TradeStation who want equities, options, and futures behind an agent |
| CCXT / crypto-exchange MCP servers | Read / trade | Crypto exchanges, not equity brokers | Self-host, broker API key | Builders automating crypto trading or tracking across exchanges with one MCP interface |
| Alpha Vantage MCP server | Data only | None (market data) | Self-host, data API key | Grounding an assistant's research in real prices and fundamentals without touching an account |
| Polygon.io MCP server | Data only | None (market data) | Self-host, data API key | Low-latency, high-volume market data feeding an analytical agent |
| Financial Modeling Prep MCP server | Data only | None (market data) | Self-host, data API key | Fundamentals-heavy research (statements, ratios, valuation) for an agent |
| Finnhub MCP server | Data only | None (market data) | Self-host, data API key | Real-time quotes, news, and fundamentals for an assistant, with a usable free tier |
| Yahoo Finance MCP servers (community) | Data only | None (market data) | Self-host, data API key | A free, zero-setup data feed for prototypes and personal projects |
Ranked by what you want to do
There is no overall number one, because the right MCP connector depends on whether you want to trade a real account, only pull data, or build your own server. Below the field is ranked inside each use-case, with the stronger fit first. Walnut appears in the read-and-trade group as the no-code, broad-coverage pick, not as an overall winner.
Best for reading and trading a real account
If you want an AI agent that can read your positions and place orders in a real brokerage, these connect to an actual account. The right one depends on which broker you hold and how much you want to build.
- Alpaca MCP server (official). The official, open-source server with the deepest agent surface (data, paper, and live execution) for accounts at Alpaca.
- Interactive Brokers (community MCP connectors). Community connectors give read and trade access to IBKR's broad global markets for users willing to run their own gateway.
- Walnut connector. The no-code option: securely connects most US brokers and works from Claude or ChatGPT, read-only by default with trades you approve. It leads its own niche (broad coverage, no self-hosting), not the whole field.
Best for market data only (research, no account)
If you only want to ground an assistant's analysis in real prices and fundamentals, and never touch an account, these data MCPs are the answer.
- Polygon.io MCP server. Low-latency real-time and historical data across stocks, options, forex, and crypto for data-intensive agents.
- Alpha Vantage MCP server. Broad prices, fundamentals, and technical indicators with a free tier to start.
- Financial Modeling Prep MCP server. Fundamentals-first: statements, ratios, valuations, and earnings for company-level analysis.
Best for developers who want to build their own
If you write code and want full control over the tools your agent exposes, the open-source, self-hosted servers are the foundation.
- Alpaca MCP server (official). Official open-source server with paper trading, so you can iterate on agent behavior before risking real money.
- Tradier (community MCP connectors). A simple REST broker with strong options support, easy to wrap behind a custom MCP server.
- CCXT / crypto-exchange MCP servers. One MCP interface across many crypto exchanges via the CCXT library when your agent trades crypto, not equities.
How we evaluated these
We split the field into brokerage connectors (which touch a real account) and market-data servers (which do not), because comparing them on the same axis is misleading. Within the brokerage group we weighed broker coverage, whether the server can execute or only read, the auth and security model, paper-trading support, and setup effort. We treat read-only defaults and credential safety as table stakes, not bonus features: a connector that can trade but mishandles access is not better than a careful read-only one. We did not crown a single overall winner; the right server depends on your broker and whether you want execution. Facts here are qualitative and point-in-time, so verify current capabilities on each project's documentation.
Are brokerage MCP connectors safe?
They can be, if the access model is sound. The Model Context Protocol was designed by Anthropic for scoped, permissioned access, and a careful brokerage MCP leans on that. The safer setups share three traits. First, read access is the default and any trade is gated behind explicit approval. Second, the server never stores your broker password: it uses scoped API tokens (Alpaca, Tradier), the broker's own session (IBKR), or a regulated aggregator (Walnut). Third, you can scope what the agent may do, so a read-only key cannot place orders even if the model asks.
The risks worth naming: a broad or mis-scoped API key can grant more than you intend; a self-hosted server you do not secure becomes a new attack surface; and community connectors vary in maintenance, so vet the code before pointing it at a live account. Paper trading first is the simplest guardrail. Walnut follows the safe pattern: your broker login and credentials stay at your broker, access is read-only unless you enable trading, and every order requires your confirmation.
Which MCP connector should you use?
The quickest way to narrow it down is to match the connector to what you actually want the agent to do.
- You hold an Alpaca account and want full agent control. Alpaca's official open-source server is the deepest fit, with data, paper, and live trading you self-host.
- You hold Interactive Brokers and want global market access. A community IBKR connector reads and trades through the IBKR gateway, for users comfortable running it.
- You want options trading behind an agent. Tradier's REST API and community MCP servers handle equity and options orders with a sandbox.
- You want broad US-broker coverage with no code. Walnut securely connects most US brokers and works from Claude or ChatGPT, read-only by default with trades you approve.
- You only want research, never account access. Pair a data MCP: Polygon for low-latency data, Alpha Vantage for breadth, Financial Modeling Prep for fundamentals.
- You trade crypto, not equities. A CCXT-based MCP server connects exchanges like Coinbase and Kraken behind one interface.
Where Walnut fits
To be upfront, since this is our site: Walnut is the AI investing assistant that talks to the broker you already have and places the trades you approve. Among brokerage MCP connectors it is the portfolio-aware, read-only-by-default consumer option, rather than a raw developer connector, and it leads in that niche rather than overall. Walnut is itself a portfolio-aware MCP connector: it lets Claude Desktop, ChatGPT, or Cursor read your real connected brokerage portfolio and, where the broker supports it, place trades you approve. You can use either Walnut's hosted connector or a small downloadable local MCP shim, whichever your AI client supports. It connects the broker you already have securely, so your broker login stays at the broker; access is read-only by default; and every trade requires your approval. Because broker feeds rarely pass cost basis, Walnut frames returns as window returns rather than realized profit and loss, and says so. As a hosted MCP connector for Claude, ChatGPT, and Codex (and any MCP client), you connect it by adding its MCP 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; it never places trades. Full per-client setup lives on the Walnut MCP connector page.
If you are a developer who lives in Alpaca or IBKR and wants full execution control, the Alpaca official server or an IBKR community connector will fit better. For pure research, pair either approach with a data MCP like Alpha Vantage or Polygon. From a connected account you can ask an assistant about a specific stock, ETF, or theme against your real holdings. For the wider field, see how to connect your brokerage to an AI assistant and the roundup of the best finance MCP servers in 2026. Walnut is not an investment adviser.
The bottom line
Most finance MCP servers serve data; only a handful read and trade a real account. For execution, Alpaca's official server is the strongest self-hosted option for Alpaca accounts, community connectors cover Interactive Brokers and Tradier, and Walnut is the no-code, broad-coverage connector that securely links most US brokers and works from Claude or ChatGPT, read-only by default. For research only, Alpha Vantage, Polygon, Financial Modeling Prep, and Yahoo Finance feed prices and fundamentals but cannot touch an account. There is no single best one: pick by your broker, whether you want execution, and how much you want to build, and keep access read-only until you deliberately enable trading.
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.
FAQ
What is the best MCP connector for a brokerage?
There is no single best one; it depends on your broker and whether you want to build. For one Alpaca account with full agent control, Alpaca's official open-source MCP server (alpaca-mcp-server) is the deepest. For Interactive Brokers, community MCP connectors read and trade through the IBKR gateway. For broad US-broker coverage with no code, Walnut is a strong pick: it securely connects most brokers and works from Claude Desktop, ChatGPT, or Cursor, read-only by default with trades you approve. Market-data MCPs like Alpha Vantage and Polygon cannot touch an account at all.
What is an MCP connector for a brokerage?
MCP (Model Context Protocol) is an open standard, introduced by Anthropic in late 2024, for connecting AI assistants to external data and tools. An MCP connector for a brokerage is a server that exposes your account to an assistant like Claude or ChatGPT as MCP tools, so it can read positions and, with the right server, place trades.
Which MCP servers can actually trade a real brokerage account?
Brokerage-execution MCP servers can. Alpaca's official server reads and trades Alpaca accounts (with paper trading). Community connectors for Interactive Brokers and Tradier read and trade those accounts. Walnut securely connects most US brokers and places trades you approve. Market-data MCPs like Alpha Vantage and Polygon cannot touch an account at all.
Is there an official MCP server for a brokerage?
Alpaca publishes an official, open-source MCP server. Interactive Brokers and Tradier do not have official MCP servers, so their connectors are community-built wrappers around each broker's API. Walnut is a hosted connector rather than a self-hosted open-source server, and it covers many brokers instead of one.
What is the difference between a brokerage MCP and a market-data MCP?
A brokerage MCP (Alpaca, IBKR connectors, TradeStation, Tradier, Walnut) connects to your actual account so an assistant can read holdings and possibly trade. A market-data MCP (Alpha Vantage, Polygon, Finnhub, Financial Modeling Prep, Yahoo Finance) only serves prices, fundamentals, and news, with no account access. Many setups pair a data MCP for research with a brokerage MCP for the portfolio.
Can Claude or ChatGPT trade my brokerage through MCP?
Yes, if you connect a brokerage-execution MCP server and grant trade access. Claude (via MCP) or ChatGPT can then place orders the model proposes. The safer pattern keeps access read-only by default and requires your explicit approval before any order. Walnut works this way: read-only unless you enable trading, and every trade needs your confirmation.
Do I need to run an MCP server myself?
For the open-source and community servers (Alpaca, IBKR connectors, Tradier, Alpha Vantage, Polygon), yes: you self-host and configure them, which needs technical comfort. Walnut is hosted, so there is no server to run; you connect a broker in the app and use the connector from Claude or ChatGPT.
Are brokerage MCP connectors safe?
The safer ones default to read-only access, never store your broker password, and scope what the agent can do through API tokens or a regulated aggregator. Before connecting, confirm whether the server can place trades and what approval controls exist. Walnut is read-only by default, keeps your broker login and credentials at your broker, and requires approval for every order.
Which MCP server connects Schwab or Fidelity to an AI assistant?
Neither Schwab nor Fidelity publishes an official MCP server, so the practical route is an aggregator-based connector. Walnut reaches both, so a Schwab or Fidelity account can be read from Claude or ChatGPT (Schwab also supports trade execution; Fidelity connects read-only today). Plaid is a regulated aggregator used for read-only account linking. Single-broker servers like Alpaca or Tradier do not touch Schwab or Fidelity at all.
Is there an MCP connector for TradeStation or Interactive Brokers?
Both have community-built connectors rather than official ones. TradeStation and Interactive Brokers each expose a full brokerage API, and open-source MCP servers wrap them so an agent can read accounts and place orders through your own gateway or OAuth session. Newer names such as Connect Trade also appear in this space. Because these are community projects, vet the code and maintenance before pointing one at a live account, and start with paper or simulated trading.
Which MCP connector works with the most brokers?
Walnut covers the most US brokers because it connects through a regulated account-aggregation layer that reaches many brokerages behind one secure link. Alpaca, Tradier, and IBKR connectors are single-broker by design. If you want one connector across several brokers without writing code, the aggregator-based approach reaches the widest set.
Can an MCP connector trade Robinhood?
Not for live trading today. Robinhood has no public trading API, so it connects read-only through regulated aggregators (Walnut can read a Robinhood account but not trade it). For execution, brokers such as Alpaca, Public, Schwab, Tradier, and Webull are supported. Verify current support before relying on it.
What is the best MCP connector for paper trading?
Alpaca's official MCP server is the strongest for paper trading: it exposes a full sandbox so an agent can place simulated orders before touching real money. Tradier also offers a sandbox through community connectors. Paper trading is the safest way to test agent behavior before granting live trade access.
Can an MCP connector trade crypto instead of stocks?
Yes. CCXT-based and exchange-specific MCP servers connect crypto exchanges like Coinbase, Kraken, and Binance for balances, market data, and orders. These are separate from equity brokerage connectors. Scope your exchange API keys carefully (read-only versus trade) because a broad key grants more access than you may intend.
Which MCP connector should a non-developer use for a brokerage?
A hosted, no-code connector. Self-hosted servers like Alpaca's or the IBKR community connectors require running and securing a server. Walnut is the no-code path: you securely connect a broker and use it from Claude or ChatGPT with nothing to host, read-only by default and with trades you approve.
Walnut is informational and is not an investment adviser. MCP servers, broker support, auth models, and pricing change quickly; verify current capabilities on each project's documentation before connecting an account. Nothing on this page is a recommendation to buy, sell, or hold any security or to use any particular product.
ETFs and stocks in this guide
Stocks: IBKR