Bring your own data: MCPs are live in Investi

Ask ten investors what they look at before buying a stock and you get ten different answers. One lives inside SEC filings and will not touch a company until the balance sheet confesses. One watches macro data the way other people watch sports. One swears the truth is in web traffic and app store rankings, weeks before it shows up in a quarterly report. One reads the crowd, because sentiment turns before prices do.

None of them are wrong. That is the whole point.

Investing is personal. The data you trust is part of your opinion, and no tool should pick it for you.

Which is exactly what most tools do. Every screener, terminal, and AI investing app ships with a fixed menu of data, chosen by someone who is not you. Whatever your process needs beyond that menu simply does not exist as far as the tool is concerned. You adapt to the tool, instead of the tool adapting to you.

As of today, Investi works the other way. MCP support is live in the beta: connect any MCP server, and its data becomes something your analyst can reach for while it works. Pick from the catalog or paste a URL, and the tools show up in seconds, right next to the market data.

Here is what that unlocks, and why we think it completes an idea we have been building toward from the start.

Your data diet is an opinion

A professional analyst can, in principle, look at everything about a company and then a little more: the filings, the call transcripts, the foot traffic, the job postings, the subreddit. But no real analyst looks at everything equally. The craft is in the diet. What you read first, what you weight heavily, and what you cheerfully ignore is where an actual investment style lives.

The fundamentalist and the sentiment reader can hold the same stock for entirely different reasons, and both can be right. Their edge is not the intelligence they apply. It is the inputs they chose to apply it to.

So here is the uncomfortable question for any AI analyst: whose diet is it on? If the product decides which data exists, the product has quietly decided how you invest. The bias is in the tool before you type a word. We wrote about this in the context of skills, and it is just as true for data: we refuse to bake opinions in. An analyst that only eats what we feed it would be our analyst wearing your name.

Bring your own opinions. Bring your own process. Bring your own data.

MCP, in one honest paragraph

MCP stands for Model Context Protocol, an open standard Anthropic released in late 2024. OpenAI and Google adopted it within months, and in late 2025 Anthropic donated the whole protocol to the Linux Foundation, so today no single company owns it. Before it existed, connecting an AI to a data source meant a custom integration for every pair of tool and source, which is how you end up with spaghetti. MCP replaces the spaghetti with a plug. A data provider runs one server, any compatible AI can call it, and everybody stops writing adapters.

If you have heard us say that skills are an open standard from the coding world that maps beautifully onto investing, this will sound familiar, because it is the same story one layer down. Skills let you hand the analyst your method. MCP lets you hand it your sources. Both are plain, portable, and chosen by you.

What it looks like in Investi

Inside the app there is now an MCPs tab. It shows the servers you have connected, each with the tools it hands the analyst, and a catalog of research servers we have already vetted: live fundamentals and filings, crypto markets, drug approvals and clinical trials for biotech watchers, web traffic and app rankings, social sentiment, global news events, prediction markets, and general web search. If your source is not on the shelf, paste its URL and connect it yourself.

From then on, no ceremony. Ask a question, and the analyst pulls from whatever you connected when it is relevant, the same way it already pulls market data. Want to be explicit? Type / in chat and point it at a specific server yourself.

And this is the part worth slowing down for: the point of connected data is not abundance, it is grounding. A raw model asked about last quarter will answer from memory, confidently, with a number that may be a year stale. An analyst with your servers connected goes and gets the actual figure, and shows you where it came from.

Three investors, three setups

To make it concrete, here is the same product shaped three different ways.

The fundamentals purist connects filings and financial statements and nothing else. Their analyst reads businesses the old fashioned way: revenue quality, margin trends, insider buying, what management said versus what they did. Sentiment never enters the room, because the purist never invited it.

The biotech specialist connects clinical trial registries and drug approval data next to the market feeds. When a phase three readout lands, their analyst can connect the science to the position in one pass. Try asking a generic chatbot to do that from memory.

The signal hunter connects web traffic, app rankings, and social listening. Their analyst watches the quarter unfold in near real time and flags when the alternative data disagrees with the narrative. When the crowd and the download numbers point in opposite directions, that is exactly the kind of tension worth investigating.

Same harness, three completely different analysts. That is not a feature list. That is the philosophy doing its job.

The trilogy is complete

We keep coming back to one line: we build the harness, the opinions are yours. Skills were the first act of that promise. You could hand the analyst your method, your checklists, your way of reading a business. But your method was still running on our choice of data, and that always felt like an asterisk.

MCPs remove the asterisk. Now the workspace is yours, the process is yours, and the inputs are yours. The analyst starts neutral and becomes exactly as opinionated as you make it, in exactly the direction you choose.

We build the harness. The engine keeps getting better. And now you choose the fuel.

MCPs are live in the beta today. Connect your first server from the features page, and if you want the deeper story on why the layer around the model is the real product, start with Anatomy of a harness.