AI Assistant Connector

Trandence speaks MCP (Model Context Protocol), the open standard AI assistants use to reach outside data. Connect it once and your assistant can read your trades, statistics, setups, playbooks and journal directly — no exporting CSVs, no pasting screenshots, no copying numbers by hand.

Ask “how did I trade in February?” and the assistant reads the real figures out of your journal instead of guessing from what you told it.

Before you connect

You need a Trandence account with your trades imported — the connector reads your journal, so an empty one has nothing to say.

You do not need to be signed in first. If you aren’t, the connection takes you through sign-in and brings you straight back to the approval step.

Connecting Claude

Works on every plan, including Free — though a free account is limited to one connector in total. The steps are the same in the browser and in the desktop app.

1. In Claude, open Customize > Connectors, click Add, and choose Add custom connector.

2. Give it a Name — Trandence — and paste the Remote MCP server URL:

   https://api.trandence.com/api/mcp

3. Leave Advanced settings alone. The OAuth client ID and secret are optional and you do not need them: Claude identifies itself automatically. Filling them in will break the connection.

4. Click Add.

Claude opens a Trandence page asking you to authorize access. It names the app, shows which account it will read, and lists exactly what it may do. Click Allow.

That is the whole setup. Claude will show the available tools once connected.

Connecting ChatGPT

ChatGPT supports custom MCP connectors on Pro, Max and Enterprise plans, behind a developer setting.

1. Open Settings > Apps > Advanced settings and turn on Developer mode. On Business and Enterprise workspaces an admin has to permit this first, under Workspace Settings > Permissions & Roles > Connected Data.
2. Go to Settings > Connectors and click Create.
3. Give it a name — Trandence — and enter the server URL: https://api.trandence.com/api/mcp
4. Choose OAuth as the authentication method. Leave client credentials empty.
5. Save, then authorize. You will land on the same Trandence approval page described above.

Keep the connector scoped to Individual unless you actually want colleagues reading your journal — an Organization connector is shared.

What the assistant can and cannot do

Access is read-only unless you say otherwise, and read-only is the default everywhere: the approval box is unticked, and so is the one in the key dialog. Click straight through and the assistant can look but not touch.

There is no version of this connection that can place a trade, change a price, edit or delete a journal entry, or touch your settings. That holds however you connect it.

The one thing you can allow is tagging. Tick the box and the assistant may attach your own setups and playbooks to trades you left untagged — nothing more. It cannot change a tag you set yourself; if it disagrees with one of yours it has to tell you and leave it alone. Anything it does tag is marked in Trandence as its work, so you can always tell your classifications from its, and you can change or clear any of them.

Why the option exists: playbook and setup analytics only work on trades that carry tags, and tagging months of history by hand is the job everyone puts off. This lets you talk it through instead.

It reads only your own data. Nothing about other traders is reachable through the connector.

What it can read:

• Performance summaries — P&L, win rate, profit factor, fees, over any date range
• Day-by-day and per-ticker breakdowns
• How each playbook actually performs — so “I’m planning to trade this setup” can be checked against its record
• Results by setup grade, next to the risk matrix that says how much each grade may risk
• Individual trades, including fills and your playbook notes
• Discipline metrics, with the sample size each score is based on
• Your daily journal, and a search across it
• Your own setups and playbooks

Those middle two are what make the connection worth having. The assistant grades your trades against your rules, not against generic trading advice — and it can tell you whether the grades you size up on are the ones that actually pay.

They also depend entirely on you: a trade with no playbook attached and no setup grade can’t be measured that way. If most of your trades are untagged, start there — it’s the difference between an assistant that reports numbers and one that can judge your process.

Things worth asking

Start here — these use the connector well:

• “Analyse my trading for last month.”
• “Which tickers do I overtrade, and what does it cost me?”
• “Was I disciplined last week? Be honest.”
• “Grade my last ten trades against my own setups.”
• “What is my most frequent recurring mistake? Look across the whole journal.”
• “How much of my P&L is going to commissions?”
• “Find every day I wrote about revenge trading and show me what happened on those days.”

Two habits that make the answers much better:

Ask for a specific period. “February” or “the last two weeks” beats “recently” — the assistant fetches exact figures for a date range.

Ask it to check your rules, not just your results. A green day with a broken process is still a bad day, and the connector gives the assistant everything it needs to say so.

Numbers you can trust

Every figure comes pre-computed from Trandence itself. The assistant is not adding up trades and doing arithmetic — it receives the same P&L, win rate and R-multiples the app shows you, so the two can never disagree.

Where data is missing, the connector says so explicitly rather than filling the gap. If you did not record which rules you followed, discipline shows as no data — not as a score of zero. A month with no trades returns “nothing recorded”, not a neutral-looking number that reads like a finding.

Managing access

Open Settings > Integrations in Trandence to see what is connected and to disconnect anything you no longer use. Revoking takes effect immediately.

The same screen issues API keys for tools that connect with a token instead of a browser sign-in — command-line assistants and IDE integrations, typically. A key is read-only unless you tick the tagging box while creating it; keys that carry it are labelled in the list, and every key is shown once when created.

Other assistants

The connector is an ordinary remote MCP server, not something built for one vendor. Anything that speaks the standard can use the same URL: it is served over HTTPS with Streamable HTTP transport and OAuth, which is the combination assistants ask for.

Command-line and IDE assistants usually want a config file rather than a browser sign-in. Those take an API key from Settings > Integrations instead, sent as a bearer token:

{
  "mcpServers": {
    "trandence": {
      "type": "http",
      "url": "https://api.trandence.com/api/mcp",
      "headers": { "Authorization": "Bearer YOUR_KEY" }
    }
  }
}

Treat that file the way you would a password: the key in it grants read access to your journal until you revoke it, and it does not belong in a git repository.

Support for MCP differs between products and moves quickly. If the steps above no longer match what you see, look for “custom connector”, “remote MCP server” or “developer mode” in your assistant’s settings — the server address is the only thing it needs from us.