Once you have a market data source, there are two ways to put it to work. Write a small program that fetches the same thing on a schedule. Or connect an agent and ask it questions. People often reach for the agent because it is new. Many jobs are better done by the script.
What a script is good at
A script makes the same request and handles the same response every time. Fetch the dealer levels. Write the walls and the time to a file. Stop. It will do that a thousand times without changing its mind about what a wall is.
NoVo’s free data is available as plain GET requests for exactly this kind of use. /api/levels, /api/vol and /api/calendar return JSON with no key. There is no envelope to unwrap and no model in the loop. The choice between the two doors is covered in MCP or a REST API.
A script also fails in the open. If the response is an error, the script stops or logs it. It does not write a plausible number in its place.
What a script cannot do
It cannot answer a question you did not plan for. It cannot read two payloads and say how they relate. It cannot explain a term. Every new question means more code.
What an agent is good at
An agent takes a question in plain words and decides which tools to call. It can gather several readings, set them side by side and write a short page. It can answer the follow-up. For work that changes shape from day to day, that is worth a lot.
What an agent costs you
It may not do the same thing twice. Asked the same question on two mornings, it may call different tools or lay the answer out differently. It can drop a caveat. It can fill a gap from memory if its instructions allow it. Each of these is manageable, and each needs watching. A script needs none of that attention.
There is also the plain matter of requests. An agent exploring a question may make several calls where a script would make one. That is fine for a question and wasteful for a routine, as set out in an agent built around rate limits.
Sorting the jobs
Ask one thing: does the job change? Recording the levels every morning does not change. That is a script. Reading this morning’s levels against the calendar and telling you what stands out does change. That is an agent.
A second test is what a mistake looks like. If a wrong output would go unnoticed for a week, prefer the script, because its mistakes are loud.
Using both
The strongest setup uses each for its strength. The script collects on a schedule and keeps a clean file of readings with their times. The agent reads that file when you have a question, and calls live tools for anything current. The record is exact because a program wrote it. The reading is flexible because a model did it. Why a kept record matters is in one reading tells you where, a series tells you which way.
Both doors are on the MCP & API, and they serve the same readings with the same freshness stamps.