StampedeStampedeStampede home

Trading MCP: Connect an AI to Your Trading Account

What a trading MCP is

MCP, the Model Context Protocol, is an open standard that lets an AI assistant talk to an outside system through a defined set of tools. An MCP server exposes those tools; the AI calls them. That is the whole idea. It is the same plumbing behind connecting Claude to a codebase, a database, or a calendar.

A trading MCP applies that plumbing to a trading account. Instead of reading a screenshot or a pasted chart, the AI gets real tools: read the account, list open positions, check the rules that apply to it, place or close an order. It works against your account state, not a description of it. That is the line between an AI trading assistant that comments on charts and one that is actually connected to the account.

Stampede runs on the Match-Trader engine, and the connector sits on top of it. So an AI can read and act on your account through the Match-Trader API without you writing any integration code. Your account is simulated through evaluation and after funding, which matters here for a specific reason: the firm rules are enforced in code as the AI trades, not reviewed after the fact at payout. Position limits and hold requirements are checked at the point of the order. If an action would break a rule, it does not go through. The exact numbers live on /rules and how the engine is wired lives on /platform.

This is the first prop firm you can connect an AI to, and the first MCP built for a prop-trading account rather than a personal brokerage one. You connect it with your own login, and you can revoke access whenever you want; the full scoping and control model is in the safety section below.

Charcoal screen-print of an American bison with fine ember-orange circuit traces glowing across its shoulder, representing an AI connected to a trading account.

MCP, API, and bot or EA: three ways to connect

There are three ways to put software between your intent and your trades. They are not interchangeable, and the difference decides how much you build, how much control you keep, and when trades actually happen.

An MCP server is the newest of the three. MCP (Model Context Protocol) is an open standard that lets an AI assistant read and act on an outside system in plain language. A trading MCP connects an assistant like Claude to your account. You ask it to check your open positions or place an order, and it does, through a connector you authorize with your own login. You stay in the conversation, and nothing happens unless you ask for it.

An API is the layer underneath. A forex trading API is the interface a developer writes code against, and you or someone you hire builds the logic, handles the authorization, and maintains it. It is the most flexible option and the most work. Point that same interface at an evaluation or funded account instead of a personal brokerage and you have a prop trading API.

A bot or EA runs on its own. An Expert Advisor (EA) executes a fixed rule set without you present. An AI trading bot is the same pattern with a model making the calls instead of a static rule tree. You set it running and it keeps going, so Match-Trader automation through a bot trades whether or not you are watching.

ApproachWhat it isYou need toRuns when
MCPAn AI assistant connected to your account in plain languageAuthorize a connector with your own loginYou ask, inside a conversation
APIThe raw interface code is written againstWrite and maintain the code yourselfYour code calls it
Bot / EAA pre-set strategy that trades on its ownConfigure or build the strategyContinuously, unattended

On a Stampede account, all three point at a simulated evaluation or funded account rather than a personal brokerage, and the firm rules are enforced in code as you trade instead of reviewed after the fact. See /rules and /platform for how that works.

How you connect an AI to your account

The connection runs on MCP, the open standard AI assistants use to talk to outside tools. Four steps, in order:

  1. Open the connector from your dashboard. It lives on the platform, next to your account.
  2. Authorize it with your own account login. You grant the access, and nothing connects without you.
  3. The AI gains a scoped set of actions, limited to your account and nothing else.
  4. Talk to it in plain language. No code, no config files.

Once connected, the assistant works inside the boundaries you set. Here is what that looks like in practice:

You askThe AI does
"What are my open positions?"Reads your positions and current account state
"How close am I to my daily limit?"Answers questions about your account
"Close half my EURUSD position"Helps place or manage trades, within limits you set

This is how you connect an AI to a trading account without handing your credentials to a third party. The connector only touches your own account, and you can revoke access whenever you want.

It works with Claude and ChatGPT-style assistants that support MCP. Claude trading through the connector reads the same account data you see on screen, so you and the assistant are always looking at the same numbers.

One thing to be clear about. An AI trading assistant helps you see and act faster. It does not predict markets, and it does not guarantee anything. The value is speed and clarity on your own account state, not a forecast.

When you connect an AI to a prop account, whether an evaluation or a funded simulated account, the firm rules stay enforced in code as you trade. See /platform for the actual connect step, and /rules for exactly what is allowed.

The prop-account difference

Almost every trading connector you can find today points at a personal brokerage account. You link a stock or crypto account you already own, and the AI reads prices and places orders against your own money. That is a real use case, and there are decent tools for it.

A prop account is a different thing. You pay a fee to take an evaluation on a simulated account, you trade to a set of firm rules, and if you pass you trade a funded account on a split. The account is not your personal brokerage login. It belongs to the program, and the program has rules about how it can be traded.

That distinction changes what the connector has to do. On a personal account, the AI only has to answer one question: do you want this trade or not. On a prop account, every action also has to sit inside the firm's rules. Position size, when you can hold, how much you can lose in a day, how the account draws down. Those rules do not disappear because a model placed the order instead of you.

This is where connecting an AI to a prop account gets interesting. The rules travel with the account, so they apply to the AI exactly as they apply to you. We run on Match-Trader and add the connector on top, and the firm rules are checked in code as the order goes through, not reviewed after the fact at payout. If a trade would break a rule, you find out when it happens, not weeks later when you go to withdraw. You can read the exact rules on our rules page and how the account works on the platform page.

That is the part no one else has built. Stampede is the first prop firm you can connect an AI to, and this is the first MCP for a prop-trading account. The connectors that already exist point at brokerage accounts with no rules layer to respect. A prop account has one, and here it is enforced while you trade rather than checked at the end.

Why rules-as-code changes the trust math

Most prop accounts run on trust that shows up late. You trade the whole evaluation, or the whole funded month, and only at payout does anyone check whether you stayed inside the rules. Break one without realizing it and you learn about it after the fact, when it costs the most. The rulebook sits in a document that a person reads later and interprets.

Rules-as-code moves that check to the moment of the action. The same firm rules a reviewer would apply at payout are written into the connector, so they run while you or your AI trade. You are not trusting someone to read the rulebook correctly after the fact. You are watching the rulebook execute as it happens.

This matters more once an AI is placing the orders. A model can act faster than you can watch a screen, and a rulebook that lives in a PDF cannot check a trade a model opened a second ago. Enforcement written in code is the only kind that keeps up with an automated trader.

At a category level, the rules fall into a few types, and the connector treats each one differently:

Rule typeWhat it guardsHow the connector treats it
Hard limitsThe account's ceiling on exposure and lossEnforced at the point of action, so the disallowed thing does not happen
Conduct rulesHow a position is held, not only whether it is openedApplied to the outcome of the trade, never to your ability to close it (the operational detail is in Running bots and EAs)
Account-state rulesWhere you stand across the full evaluation or funded periodTracked live against your own account, not reconstructed at payout

The connector does not invent rules of its own. It enforces the ones you already agreed to. The exact thresholds are the same numbers published at /rules, and how they map onto the Match-Trader platform is set out at /platform.

The result is a smaller gap between what you think is allowed and what actually is. You do not have to hold the firm's judgment in your head, or bake it into your AI's prompt and hope it sticks. The judgment is the code, and the code reads the same for everyone.

Running bots and EAs on the Match-Trader API

Automated trading is allowed. You can run an expert advisor, your own script, or an AI through the connector against the same Match-Trader API. Nothing about connecting a model is a special case; it is another client sending orders.

That is the point worth being clear about: the account does not care what sends the order. A manual click, an EA, a cron job, and a prompt to an AI trading assistant all hit the same rules-as-code layer, and the layer treats them identically. There is no separate rulebook for humans and machines, and no soft "flagged for review" tier that a bot slips into. Whatever you connect trades under the exact rules you agreed to at purchase.

The connector rate-limits requests. A runaway loop or a model stuck in a retry cycle cannot machine-gun the API or spam orders against your account, because the connector caps how fast calls go through. This is a guardrail on the automation, not on your trading: normal order flow is well inside the limit.

Automated flow stays legible after the fact. When a trade breaks a rule, it shows in your history struck through within seconds of the close, so a bot running unattended overnight does not hide anything from you the next morning. If a position is still inside a minimum hold, the interface shows the hold still counting down rather than letting a script close early and quietly void the result. You get the same record whether you placed the trade yourself or a connected agent did.

None of this is a claim about how a bot will perform. Automation is a capability: it decides how orders reach the account, not whether they make money. What Stampede adds on top of Match-Trader is that the firm rules are enforced in code as the bot trades, instead of reviewed after the fact at payout.

Safety, scope, and control

Start with what the connector cannot reach. It touches your trading account and nothing else. It cannot request a payout, change your payout address, or move a single dollar out of the account. It cannot see or edit your personal details, your identity documents, or your billing. It cannot reach another trader's account, and it cannot touch any firm system behind the account. Its entire world is one account: the one you point it at.

You authorize it with your own trading login. There is no separate master key and no shared credential. The AI acts as you, inside the account you already control, and only for as long as you allow it.

Revoke access whenever you want. When you revoke, the connection stops. The AI cannot reconnect on its own, and nothing it set up keeps running once access is gone. You do not have to explain the decision or wait on a support queue to make it.

Inside that scope, the actions split into two kinds. Reading is the connector pulling your positions, balance, and market data so the AI knows the current state. Writing is the connector placing, adjusting, or closing a trade. Both stay inside the one account, and both are bounded by the account's own limits.

Those limits are the firm rules, and they run in code while you trade rather than getting reviewed after the fact. The connector does not sit outside the rulebook. It operates under the same evaluation and account rules a human trades under, checked as the order goes through. If an AI action would break a rule, it is handled the same way the platform handles it for a person. The specifics of each rule live on /rules, and how the account itself works is on /platform.

The account is a simulated evaluation or funded account, so the AI is trading in a simulated environment, not against your bank. That is a deliberate design choice, not a limitation. You get to connect an AI, watch how it behaves, and keep full control of the account it runs on, with the rules doing the guarding in real time.

How to get started

Three steps.

  1. Pick an evaluation and pay the fee. Guest checkout, no account required to buy.
  2. Get your Match-Trader login when the account is provisioned.
  3. Connect your AI with the MCP connector, using that login to authorize it.

From there you trade the way you already planned to, with your AI in the loop. The one thing that changes: the firm rules run in code while you trade, not at payout. See exactly how the connector works on the platform page, and read the rules it enforces on the rules page.

Frequently Asked Questions

What is a trading MCP?

It is a connector built on the Model Context Protocol, the open standard that lets an AI assistant like Claude or ChatGPT use an outside system through a defined set of actions. A trading MCP applies that to a trading account, so the assistant can read your account and, within limits you set, help you act on it. It is a capability, not a strategy or a promise of results.

Can I connect an AI to my trading account?

Yes. You authorize the connector with your own account login, the assistant gains a scoped set of actions, and you talk to it in plain language about your account. On Stampede this connects to a prop account running on the Match-Trader engine. See /platform to connect.

Is there an AI agent that can trade for me?

An AI assistant connected over MCP can place and manage trades within the limits you grant it, and bots or EAs can run a strategy through the Match-Trader API. You set the boundaries in both cases. Automation does not change the rules and it does not predict markets; the same rulebook applies whether a person or code sends the order.

Is it safe to connect an AI to my account?

You authorize the connector with your own login, it only touches your own account, and it is limited to the actions you allow. You can revoke access at any time, and access ends when you do. The connector does not move your money.

What is an API in trading, and how is it different from an MCP?

A trading API is the raw programmatic interface a platform exposes for reading data and sending orders. An MCP wraps that API so a natural-language AI can use it directly. Developers tend to use the API; an MCP suits someone who wants to work with an AI assistant against a live account.

Can I use Claude or ChatGPT for trading?

Yes, with assistants that support MCP. Once connected, the assistant can read your account and help you act within the limits you set. It helps you see and act faster; it does not forecast markets or guarantee outcomes.

Does connecting an AI or a bot break prop firm rules?

On Stampede, automated and API trading is allowed within the same hard rules that apply to manual trading. Because the connector runs on the same engine that enforces those rules, an action that would break a rule is caught as you trade rather than reviewed later. The full rulebook is at /rules.

Follow the herd.