MMT, in Cursor.
Your agent should see the same market you do. Not a PDF of the docs.
Pasting docs into ChatGPT is not integration.
You already write scripts with an LLM. Today that means copying llms.txt, praying the model remembers BOOK vs OHLCV, and pasting the broken bit back into the terminal. That is homework.
MCP is the opposite. The agent sits next to the chart. It can see MMT context and help you write the indicator without the copy-paste loop.
Scripting is code that runs on the chart. MCP is the agent sitting next to it. Different surface. Both are the point.
Stay in the editor.
Cursor, Claude, whatever you already live in. Ask what the book is doing. Ask for a delta histogram on aggregated perps. Get a script back that uses our functions, not a hallucinated Pine port.
Same data the terminal uses.
Books, heatmaps, volume delta, liquidations. The same feed the chart is on. We are not bolting a chatbot onto a candlestick and calling it AI.
This is not a trading bot.
It does not place your orders. It does not DM you signals. It makes the terminal legible to the tools you already use to write code. You still read the tape.
Ask the book from the editor you already have open.
Your agent should see the same market you do. Not a PDF of the docs.