MCP + on-site agent · read-only v1

Connect AI to Your Trading Journal (MCP + On-Site Assistant)

Ask questions about your trades, win rate, and journals using bring-your-own LLM tools (MCP) or TradeJournal's on-site agent—one data layer, two ways to connect.

Cursor MCP setup Connect AI (token) Start free trial
TradeJournal Connect AI settings with personal MCP token management

Quick summary

  1. Connect AI: Create a personal MCP token at Account Settings → Connect AI (Pro / AI Assist).
  2. Set up Cursor: Create a token in Connect AI, then follow the Cursor MCP setup guide (mcp.json + localhost gateway).
  3. Start with context: Tell your AI to call get_runtime_context first, then ask about stats or trades.

Cursor mcp.json server URL: https://tradejournal.co/mcp (protocol endpoint). This HTML guide lives at /mcp-guide/.

Key takeaway

TradeJournal.co offers an AI trading journal MCP connector and on-site assistant that read only your account's trades, stats, and published posts using the same secure tool layer; generate a personal token under Connect AI for external LLMs, or use the in-browser agent with your login session (14-day free trial and included pro trader onboarding call with subscription).

MCP vs on-site agent vs spreadsheet

Surface Auth Best for Data scope
External MCP (BYO LLM) Personal token (Connect AI) Cursor, Claude Desktop, automation in your IDE Your account only; read-only v1
On-site agent Django session (login) Quick questions in the browser; no mcp.json Same tools via mcp_services; no token in JS
Spreadsheet export Manual export Static analysis; no live AI Stale the moment you export

Spec reference: Model Context Protocol. Engineering details: internal gameplan (not duplicated on this marketing page).

Choose your AI app

Most traders start with Cursor: create a token in Connect AI, paste it into mcp.json, and connect on localhost. Other clients below; no MCP install? Use on-site AI chat.

Cursor Recommended

Default path: Connect AI token → mcp.json → localhost MCP gateway.

Setup guide

Claude Desktop

Desktop MCP server config.

Setup guide

Windsurf

Similar to Cursor family.

Setup guide

ChatGPT

Remote connector when OAuth ships; use on-site agent until then.

Setup guide

Gemini

Document supported path or web agent.

Setup guide

Perplexity

Discovery + MCP spec links.

Setup guide

Anti-Gravity

Confirm official MCP docs before publish.

Setup guide

Read-only tools (v1)

Always ask your AI to call get_runtime_context first. Tools never access other users or Django admin data. Full tools reference (parameters).

Tool Purpose
get_runtime_context User portfolios, active portfolio, plan flags, privacy rules. Call this first in every session.
list_portfolios Portfolio id and name list for disambiguation.
list_trades Filtered trade list (portfolio, symbol, status, limit). User-owned rows only.
get_trade Single trade by id (must belong to the authenticated user).
get_trade_stats Aggregate win rate, profit, and related stats for the user.
get_dashboard_stats High-level public/private trade and post rollups.
list_posts Published posts metadata for the user.
get_post One published post by id.
get_portfolio_stats Per-portfolio performance rollup.

Get your MCP token (Connect AI)

Pro / AI Assist subscribers can generate personal tokens in account settings. Tokens are shown once—store them in your MCP client only.

TradeJournal MCP token created screen with copy-once warning
TradeJournal on-site AI assistant answering a question about trade statistics

Open Connect AI Free onboarding call

Example mcp.json (placeholders only)

{
  "mcpServers": {
    "tradejournal": {
      "url": "http://localhost:3333/mcp",
      "headers": {
        "Authorization": "Bearer tj_mcp_u_<YOUR_TOKEN>"
      }
    }
  }
}

Use http://localhost:3333/mcp on localhost after the MCP gateway is running. Never point dev tools at production MCP.

Related guides & journal resources

Frequently asked questions

TradeJournal MCP is a read-only bridge so external AI tools can query your own trades, portfolios, stats, and published posts via the Model Context Protocol—without exposing other users' data.

The on-site agent uses your TradeJournal login session and the same backend tools; you do not paste a token into the browser. MCP is for IDEs and desktop AI apps.

No. Every tool resolves the authenticated user from your token or session only. User id is never taken from tool arguments for authorization.

v1 is read-only. AI cannot place trades, edit posts, or change account settings via MCP.

Start free trial — claim your onboarding call

Connect AI, import trades, then ask your journal real questions—not generic chatbot noise.

Start free trial Consulting