AI Agent + MCP Trading Bot Workflows: Read, Paper and Permissioned Control

An MCP trading workflow should give an assistant the right context without turning a chat into an unchecked trading engine. With Wiz by TradingWizard, compatible clients can read public market context and, after OAuth, access the signed-in user's permitted bot and paper-workflow context. Read access, paper-write access, and portfolio-write access use separate scopes. TradingWizard does not place real-money orders.

The practical sequence is simple: connect the assistant, read the current market or bot state, inspect the reason and risk context, and use a clearly scoped paper action only when the user has authorized it. Every real-money decision remains outside TradingWizard.

What TradingWizard MCP is for

The public endpoint is https://www.tradingwizard.ai/api/mcp. It uses Streamable HTTP. TradingWizard's public MCP documentation lists public read-only tools for market data and public bot context, while private tools use OAuth and follow the account's plan and permissions.

Use the connector to ask grounded workflow questions such as:

• What is the current market price and macro context?
• What is my bot's current state and most recent timeline event?
• Did a scan return WAIT, AVOID, DATA STALE, or a tradeable setup?
• What relevant reason, level, or risk note can I inspect as the bot owner?
• Which paper positions or pending paper proposals need my review?

It is not a promise that an assistant will predict markets, make a guaranteed return, or execute a real order.

Ready to start with a bot instead of a generic chat prompt? Choose a supported asset and deploy a paper-mode bot in Wiz, then connect a compatible assistant when you need context around that workflow.

A safe agent-to-bot workflow

The valuable part of MCP is not making more decisions faster. It is putting the right current context in front of the person reviewing a decision.

The exact tools available depend on the client's connection, the user's account, the relevant OAuth scope, and the current plan. If a tool is unavailable, do not work around its permission boundary by sharing credentials or asking an assistant to improvise.

How bot context stays useful

A generic assistant can write an explanation, but an MCP-connected workflow can inspect current TradingWizard context when the user has granted access. A bot timeline may include scan receipts and lifecycle events. Depending on ownership and entitlement, it can surface a current state, a plain-language reason, and setup levels when available.

That makes three habits especially important:

1. Ask for state before an opinion. Start with the asset, timestamp, bot state, and latest reason.
2. Preserve no-trade outcomes. WAIT, AVOID, EXPIRED, DATA STALE, and RISK BLOCKED can be the safest valid answer.
3. Keep paper results labeled. A simulated position, result, or proposal does not represent a real order or a promise about future returns.

For a step-by-step review of those fields, see How to Test an AI Paper Trading Bot. For the wider safety checklist, see How to Evaluate an AI Trading Bot.

Permissions and paper controls

MCP permissions should be specific enough that a user can tell what an assistant may read and what it may change. TradingWizard documents separate OAuth scopes for read, paper-write, and portfolio-write access. OAuth tokens and API keys should not appear in tool results.

For bot workflows, a state-changing tool can still be appropriately limited. For example, an assistant may be able to pause, resume, edit, or request a scan for the user's own bot if the applicable permission and plan allow it. A proposal flow can also require an explicit approval before a paper position is opened; if the market has moved outside the defined entry zone, the proposal should expire instead of opening anything.

Never put exchange passwords, broker credentials, recovery codes, or unrelated private data into an MCP prompt. Revoke access from the connected client's settings if a device, client, or token is no longer trusted.

What MCP does not change

MCP can make an assistant more informed. It does not remove market risk, data limitations, or the need for a human decision.

• It does not guarantee a correct market call.
• It does not turn a confidence value into certainty.
• It does not make simulated paper outcomes equivalent to live trading.
• It does not authorize a real-money order through TradingWizard.
• It does not justify bypassing an OAuth scope, plan limit, or explicit confirmation step.

TradingWizard bots use fake money against real market data. TradingWizard cannot place, modify, or close a real order. Users make their own real-money decisions and place any real orders outside TradingWizard.

For the latest endpoint, client setup, and tool details, use the TradingWizard MCP page and full MCP documentation. Those public documents are the source of truth when a client or tool capability changes.

Bottom line

Use MCP to inspect current market and bot context, not to hand control of your money to a chat window. Keep reads and paper actions scoped, review the bot's state and reason before acting, preserve no-trade outcomes, and keep every real-money decision outside TradingWizard.