Claude for Stock Research: MCP and Paper Review

How to trade stocks with claude: AI trading tools organize data or propose decisions; their output can be wrong even when it sounds confident.

How to Trade Stocks with Claude

The short version: install a local MCP bridge, sign in once via OAuth, then ask Claude to check your portfolio, analyze a ticker, or place a paper-trade order. With Tradewink, Claude becomes a chat interface to your research and a paper-trading account without writing a single line of code — Tradewink's public offering is paper trading only.

This guide walks through the full setup using Tradewink's hosted MCP server. The same pattern works with Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, and any other client that speaks the Model Context Protocol. If you are comfortable pasting a JSON snippet, you can be paper trading through Claude in roughly 30 seconds.

What Is MCP and Why Does It Matter for Trading?

The Model Context Protocol (MCP) is an open standard from Anthropic that lets AI assistants connect to external tools through a uniform interface. Instead of a bespoke integration for each data source, a single server exposes tools (get_quote, place_order, analyze_ticker) and any MCP-compatible client can call them.

For trading, MCP matters because it closes the loop between natural-language reasoning and broker execution. Your AI can look at a chart, score a setup, and place the order — all in one conversation. Before MCP this required custom code, API keys stored in plaintext, and a fragile integration per client. MCP makes it pluggable.

The Two Paths to Trading Stocks with Claude

There are two honest options as of September 2026 for wiring Claude to a US brokerage.

Path A: A Broker's First-Party MCP Server

Public.com ships an official hosted MCP server that connects Claude, ChatGPT, or Perplexity directly to a Public brokerage account. You authorize the connection, Claude gets tools like place_order and get_account, and you can trade stocks, ETFs, options, and crypto by chat. The catch: your account must be at Public.

Path B: A Broker-Agnostic MCP Layer

Tradewink ships an MCP server for research and paper trading. It is paper trading only: orders go to Tradewink's built-in simulator or a connected paper/sandbox broker account (for example Alpaca paper), and public plans do not include live order submission. Claude (or Cursor or VS Code) can then run research and place paper orders through the same natural-language tools.

Path A is the route if you want Claude to place real orders in your own Public account. Path B fits if you want to practice the workflow on paper, with no real money involved. The rest of this guide uses Tradewink as the example; Path A works the same way with Public's documentation.

Prerequisites

You need three things:

• A Tradewink account (free tier is fine)
• Optionally, a broker paper account (for example Alpaca paper trading; the simulator comparison covers Alpaca's paper API alongside other practice accounts)
• Claude Desktop, Claude Code, Cursor, or any MCP-compatible client

Node.js 18+ needs to be installed because the MCP bridge runs through npx. Most developer machines already have it. If you don't, install Node from nodejs.org.

Step 1: Create a Tradewink Account

Go to tradewink.com and sign up with email, Google, Apple, or Discord. The free tier gives you AI trading signals and Discord alerts; you do not need a paid plan to try the MCP flow. Finish the onboarding and land on the dashboard.

Step 2: Connect a Paper Account (Optional)

Under the dashboard, open the broker panel and add your broker's paper or sandbox credentials (Alpaca uses manual API keys, not OAuth). Public plans do not include live order submission; without a paper account, Tradewink's built-in simulator is used. Keys are Fernet-encrypted (AES-128-CBC) with a random salt per record before storage. Once the connection test succeeds, the broker shows up as "connected" in the dashboard.

Step 3: Install the MCP Bridge in Your AI Client

Add the Tradewink MCP server to your client by pasting one JSON snippet. No API key is required — the bridge opens a browser window on first launch for OAuth sign-in.

For Claude Desktop, open claude_desktop_config.json and add an entry under mcpServers. On macOS this file lives at ~/Library/Application Support/Claude/claude_desktop_config.json. On Windows it lives at %APPDATA%/Claude/claude_desktop_config.json. The entry sets command to "npx" and args to ["-y", "@tradewink/mcp"]. Save and restart Claude Desktop.

Claude Code users can run a single command: claude mcp add tradewink -- npx -y @tradewink/mcp. Cursor and VS Code Copilot use the same JSON pattern as Claude Desktop but in their own MCP config files.

Step 4: Authorize via OAuth

The first time Claude calls a Tradewink tool, the bridge opens a browser window at tradewink.com/mcp/authorize. You see a consent screen listing what the MCP client is requesting — read portfolio, place paper orders, update watchlists, and so on. Click Authorize and Claude is now connected to your account.

Tokens are cached locally in ~/.tradewink/mcp/ and refresh automatically. Access tokens live for one hour; refresh tokens live for 60 days and rotate on each use. You can revoke access any time from the Tradewink dashboard.

Step 5: Start Trading by Chat

Once connected, ask Claude anything about your paper account. Practical examples:

• "What are my open positions right now?"
• "Check my buying power and flag any margin concerns."
• "Analyze NVDA with a full multi-agent report — bull case, bear case, meta review."
• "Show me today's top market movers with the highest AI conviction scores."
• "Add TSLA and AMD to my watchlist."
• "Place a paper-trade buy on 5 shares of AAPL with a 2% stop below entry."

Claude reasons about the request, calls the right MCP tool, and returns a structured answer. Every order is a paper order — Tradewink's public offering is paper trading only — and every MCP order still requires confirmation in the client.

What Tools Are Available?

The Tradewink MCP server exposes 39 tools organized by category. Account tools return positions, buying power, and day P&L. Market data tools return quotes, batched quotes, historical bars, news, analyst ratings, earnings, and technical analysis. Watchlist tools let Claude add or remove symbols from your watchlist. Trading tools place, close, and cancel paper orders. Strategy tools return your active strategies and sell confidence scores. AI analysis tools run multi-agent research on any ticker. Scheduling tools let Claude book a recurring analysis.

Developers can see the full tool catalog at tradewink.com/developers.

Safety: What Can and Cannot Happen

A few guarantees worth internalizing:

• Public subscriptions are paper-only; separately approved private beta accounts may submit live broker orders.
• Every order placed through MCP requires confirmation in the client — Claude cannot silently submit.
• Tokens are scoped to your Clerk identity and to the specific MCP client that authorized them. A different device or a different client starts a fresh OAuth flow.
• Revoking access at tradewink.com/dashboard/api-keys invalidates the client's tokens immediately.
• Tradewink does not hold your money. Orders are paper orders in the simulator or your paper account.

None of this replaces your own judgment. Claude can be confidently wrong, market conditions can shift mid-trade, and AI analysis is a research aid, not an oracle. For public-plan users, any real trade is your own decision, placed by you at your broker.

Going Beyond Chat: Autonomous Strategies

The chat workflow is half the story. Tradewink's Paper Autopilot also runs day-trade strategies that screen candidates, score each setup with multi-agent AI, size positions by risk, and place paper orders in a simulator or a paper/sandbox account. The chat interface is how you inspect that work; Paper Autopilot is how it paper trades between conversations.

If you want Claude to change day-trade preferences, you can ask it to update risk caps (for example a maximum daily loss and a trades-per-hour limit). Claude can call the update_preference tool. Tradewink's public offering is paper trading only, and no chat command can turn on real-money trading.

Troubleshooting

If the bridge fails to open a browser, the console prints the authorization URL — paste it manually. If authorization redirects but the client does not pick up the code, confirm that port 33418 on 127.0.0.1 is not firewalled. If tokens stop working after a week, delete ~/.tradewink/mcp/ and restart the client to re-authorize. If Claude shows no Tradewink tools after setup, check the MCP server list in Claude Desktop's settings and verify the "tradewink" entry shows as connected.

For fallback scenarios (CI, sandboxes, headless machines) set the environment variable TRADEWINK_API_KEY to a manual key from tradewink.com/dashboard/api-keys — the bridge skips OAuth and uses the key as a Bearer token.

Closing Thoughts

Trading stocks with Claude is no longer a demo. With an MCP server in front of a paper account, the workflow becomes conversational: ask, reason, decide, paper trade. The useful mental model is that Claude is your research analyst, and Tradewink is the paper-trading layer where you test its ideas — real-money trades stay your own decision at your broker.

Start on paper. Lean on the multi-agent research tools before trusting a single-model answer. Read the trade journal after every session. The traders who get the most out of AI tools are the ones who treat them as collaborators, not oracles.

How to trade stocks with claude: practical checks

Trace each factual claim to its source and timestamp. Freeze the model and prompt for a comparison, retain rejected ideas as well as winners, and evaluate hypothetical records separately from reconciled broker executions.

Read the CFTC AI trading advisory for the underlying definitions and limits. For a next step, use Backtest P&L Reconciliation: Audit Matched Trades and Autonomous Trading Agents: Workflow, Controls and Risks.