Most traders already ask a chatbot about their results: they export a CSV, paste it, and get a summary that forgets everything by the next conversation. An AI trading journal works the other way around. The assistant reads your real, up-to-date journal whenever you ask, so the answer to "which setup is costing me money this month?" comes from every trade you took, not from the rows you remembered to paste.

What an AI trading journal is

It is your usual trading journal (trades synced from the platform, setups, notes, screenshots, prop-firm accounts) plus an assistant that can query it in plain language. The value is not that the AI knows more about markets than you do; it is that it does the counting and cross-checking that nobody does by hand after a session.

Questions like "how much did I lose on Mondays after 11:00 with the breakout setup?" take five filters and a pivot table in a spreadsheet. Asked to an assistant connected to the journal, they take one sentence, and you can follow up: "and what if I exclude the FOMC weeks?".

Why pasting a CSV is not enough

  • It goes stale immediately. The file is a snapshot; tomorrow's trades are not in it.
  • It loses context. A broker export has fills and P&L, not your setups, notes, risk per trade or which account belongs to which prop firm.
  • It mixes up accounts. If you copy trades across several funded accounts, the same position appears five times and every statistic is inflated.
  • It hits size limits. A year of futures fills is thousands of rows; chatbots summarize or truncate long files, and you never know which part they skipped.

The fix is not a better prompt, it is giving the assistant a direct, permissioned connection to the journal so it can ask for exactly the data it needs: one account, one setup, one date range.

How the connection works (MCP)

MCP (Model Context Protocol) is the open standard that lets an AI assistant use external tools. A journal that publishes an MCP server offers tools such as "performance summary", "trades on a given day" or "distance to each account's drawdown", and the assistant calls them as it needs them while you talk to it.

  1. You add the server address to your assistant (in Claude, as a custom connector; in Claude Code, with one command).
  2. The assistant opens your browser, you sign in to the journal and see which permissions it is asking for.
  3. From then on, questions about your trading are answered from your own data, with the journal's numbers.

Eight questions worth asking

  1. "How close is each funded account to its drawdown right now?" The check most traders do by eye before the open, done for every account at once.
  2. "Which setup lost me the most this month, in R?" R, not dollars, so a setup traded with micros is not hidden behind one traded with minis.
  3. "Do I trade worse after a losing day?" Compare the day after a red day with your average; tilt shows up as bigger size or more trades.
  4. "Can I still request a payout under the consistency rule?" Best day against total profit, and how much is left; the consistency calculator does the same math.
  5. "How much have I paid in commissions this year?" Rarely looked at, often larger than expected on micros.
  6. "What is my best hour and my worst hour on NQ?" The trading hours guide explains why the answer differs from ES.
  7. "Tag today's trades: the first two as opening range, the rest as pullback." Journaling by voice or one sentence instead of clicking through each trade.
  8. "Is there anything big on the economic calendar during my session tomorrow?" Upcoming releases in your local time, before you plan the day.

What not to use it for

An assistant that reads your journal is good at describing what happened. It is not a signal service and should not decide entries, size or when to break your plan. Use it to find the pattern, then write the rule yourself in your risk plan.

Be careful with small samples, too. "Your win rate on Fridays is 80%" on five trades is noise. Ask how many trades a conclusion is based on before acting on it; a good assistant will tell you when the sample is too thin.

Privacy and permissions

  • OAuth, not passwords. You authorize the assistant on the journal's own page and never type your password into the chat.
  • Narrow write access. Tagging, notes and screenshots at most; no creating or deleting trades, no broker credentials.
  • Revocable. Removing the connection in the journal's settings cuts access immediately.

Frequently asked questions

Can an AI trading journal tell me what to trade?

It should not, and a good one will not try. Its job is to read your own history: which setups pay, when you break your rules, how close each account is to its limits. Entries and exits stay your decision.

Does it work with ChatGPT or only with Claude?

Jornalo exposes a standard MCP server over HTTP with OAuth 2.1, so it works with any client that supports remote MCP servers: Claude on the web and desktop, Claude Code, Cursor and others. Whether a given chatbot supports remote MCP connectors depends on that product and plan.

Can the AI place or close trades?

Not through Jornalo. The connection reads your journal and can only tag trades with setups you created, write notes and attach screenshots. It has no access to your broker and no permission to create or delete trades.