Connecting Dr. NoVo to Claude, Cursor or any AI app that speaks MCP — and how to tell whether the data coming back is worth acting on.
34 articles in this section of the Journal, grouped by topic. Every one is free to read.
Connect Dr. NoVo to your AI app 10
- What an MCP Server Is, in Plain EnglishMCP is how an AI assistant calls tools on somebody else’s server. Here is what that means, and what it does not.
- Putting a Dealer Map Inside the AI App You Already UseAn MCP connector lets an assistant read live dealer positioning directly, instead of you copying numbers into a chat window.
- MCP or a REST API: Which One You Actually WantThey deliver the same data. The difference is who is writing the code that calls it, and that changes everything.
- Putting a Market Desk Inside Your EditorThe same connector that answers questions in a chat window works where you write code. That turns out to be more useful than it sounds.
- What a Model Sees When It Calls a Market Data ToolA walk through the exact sequence, from the question you type to the sentence that comes back.
- In an Agent World, Documentation Is RuntimeA tool description is not a manual for humans. It is the thing the model reads to decide whether to call you at all.
- Read-Only Is a Promise, and It Should Be Machine-ReadableMost market data tools only look. Saying so explicitly is what lets a client run them without stopping to ask you.
- What to Ask an AI Desk About a Position You Already HoldThe useful questions are not about direction. They are about the structure around where you already are.
- What an AI Actually Does With a Dealer MapHanding a positioning map to a language model is not the same as showing it to a trader. The differences are worth knowing before you trust the answer.
- Designing a Data Feed for Something That Cannot Ask a Follow-UpHumans tolerate ambiguity by asking. An agent consumes what you sent and acts. That changes what a payload owes its reader.
Trusting what the model tells you 5
- The Number the Model Made UpLanguage models invent plausible figures. Here is how to make that failure visible instead of convincing.
- An Answer You Cannot Check Is an OpinionGrounding is the difference between an AI that reports and one that performs. It is mostly a matter of what you require of it.
- Fluency Is Not EvidenceThe most confident-sounding market analysis is often the least anchored. Here is why, and what to read instead.
- What Happens When You Ask for a Deep ReadA quick answer and a researched one are different products. Knowing which you are getting changes how much you should trust it.
- A Scan and a Search Answer Different QuestionsOne sweeps a defined universe for a condition. The other goes looking in an undefined one. Confusing them produces bad coverage claims.
Alerts over the API 6
- Standing Alerts Without a BrowserPolling a dealer map on a loop to find out when price crossed a level wastes your rate ceiling. A standing watch is the better shape.
- A Standing Alert Should Watch a Condition, Not a PricePrice alerts fire constantly and tell you nothing. The useful version watches something structural.
- The Hardest Part of an Alert Is Making It Shut UpA condition that stays true will fire forever. How a system handles that decides whether anyone keeps it switched on.
- How Do You Know an Alert Is Still Working?Silence from a monitoring system means one of two very different things, and most systems will not tell you which.
- Your Webhook Endpoint Is a Door Onto the InternetIf a service will send a POST to any URL you name, both sides have a security problem worth understanding.
- What the Signature on a Webhook Actually ProvesAn HMAC header is not encryption and does not hide anything. Here is what it does do, and why that is the useful part.
How the API behaves 4
- Rate Ceilings Are a Product Decision, Not a TechnicalityA published call ceiling is a promise about what you can build. Here is how to read one, and what happens at the edge.
- An Error Message Is Not an InterfaceIf your code has to read English to decide what to do next, the API has not finished its job.
- Your API Key Says Who You Are, Not What You BoughtKeys and subscriptions are separate systems, and confusing them produces some of the most confusing support tickets in data.
- One Key Should Open Everything You BoughtVendors who split equities and crypto into separate subscriptions are solving their own billing problem with your integration.
Data you can check 6
- How to Tell Whether Market Data Is StaleA timestamp on a payload can mean three different things. Knowing which one you are looking at is the difference between fresh data and a confident guess.
- Delayed Data and Stale Data Are Not the Same ProblemOne is a deliberate product decision you can plan around. The other is a failure nobody told you about.
- Every Number We Publish Carries Its AgeThree fields on every response, and the third one is the one nobody else ships.
- An Empty Array Is a Claim About the WorldReturning nothing when you failed to look is the most damaging thing a data feed can do, and it is almost always an accident.
- One Reading Tells You Where. A Series Tells You Which Way.Most market data is sold as a snapshot. Almost every interesting question needs the history the snapshot throws away.
- The Part of a Data Product That Cannot Be BoughtLive readings are a commodity. The archive of what the map looked like at every point in the past is not.
Buying market data 3
- A Free Tier Should Be Useful, Not a TeaserThere are two philosophies about what to give away, and they produce very different products.
- What You Are Actually Paying ForNot the numbers. The numbers are downstream of four things that cost real money to maintain.
- Ten Questions Before You Buy Market DataA short list that separates a feed you can build on from one that will surprise you in six months.