Your AI can't read behind a paywall. Here's how to connect the data it's missing.

Ask Claude who leads partnerships at ten companies, or ChatGPT what a competitor charges today, and you get a confident answer built from the free web, which is to say from nothing recent and nothing anyone paid for.

The data that would actually answer those questions exists. It's sold through APIs, behind keys and subscriptions, and none of the three apps can reach it on their own.

This guide covers what counts as paywalled data, the three ways to connect it to Claude, ChatGPT and Cursor, the exact steps for each app, and what each route costs to run.

What paywalled data means here, and what it doesn't

It doesn't mean newspaper articles. No connector gets you past a publisher's subscription wall, and anything claiming to is either scraping in breach of a license or lying.

It means data that is sold to be read by software, and that your AI app can't buy on its own:

• Company and people databases. Funding rounds, headcount by function, who holds which role, verified work emails.
• Market data. Prices, order books, on-chain activity, token metrics, as of this minute rather than the model's cutoff.
• Web-scale search and scraping. A page's full text, a site crawled whole, a hundred results for one query instead of ten links.
• Social listening. Full X and Reddit threads, hundreds of comments deep, not the three a search engine surfaces.
• Retail prices. What a product costs and whether it's in stock across Amazon, Walmart and Target right now.

Providers like Apollo, LinkedIn, X, Exa and Firecrawl sell this through APIs. It's paywalled in the sense that matters: you can't read it without paying someone, and Claude, ChatGPT and Cursor have no way to pay.

Why Claude, ChatGPT and Cursor can't reach it on their own

Each app has three sources of information, and none of them crosses a paywall.

Training data. Whatever the model read before its cutoff, usually 3 to 12 months stale on the day it ships and older every day after.

Web search. The free web, which is the same web every other user's AI reads. Search can't log in and can't pay, so you get the same answer, at the same time, as everyone else asking.

What you paste or upload. Export a CSV from the paid tool, upload it, ask your question. It works, it's manual, and it's stale by tomorrow. In Cursor it's the file you drag into context.

Every chatbot reads the same free web, so every chatbot gives the same answer. The way out is to give it a source it can query mid-answer, which is what the next section is about.

What MCP is, and why every route runs through it

MCP (Model Context Protocol) is an open standard that lets AI apps use outside tools. A server publishes tools, and the app lets its model call them while it answers. Claude calls these connectors, ChatGPT lists them under apps and connectors behind a developer mode switch, and Cursor keeps them in a config file. It's one standard with three menus.

A data connector is an MCP server whose tools return data. Connect one and your AI gains a source it can query as it writes, with no export in between.

One roundup counted about 44,000 advertised MCP servers and under 2,000 maintained ones, so the question isn't whether a connector exists for your data. It's which of three kinds you want.

Three ways to connect paid data to an AI app

Route one: the vendor's own MCP server. Apollo, ZoomInfo and Crustdata run servers for contact data; Intrinio and Fiscal.ai for financial data. You add the server as a connector, sign in with that vendor, and your AI can query your subscription. The limits are the subscription's: you see what your plan can see, one vendor per connector, one connector per app. On Claude's free plan, that's the one custom connector you get. It's the right route for a team that already pays one provider and lives in it.

Route two: your own API keys behind a custom server. This one is for developers. Wrap a provider's API in a small MCP server, or install one from a directory, and put your keys in its config. Billing stays with each vendor, and so do the keys, in every app on every machine that runs the server. It's the right route when you already pay for one provider and want full control, and the wrong one the day the third key shows up.

Route three: one connection that pays per call. Frames is an MCP server in front of a catalog of about 37,000 paid tools, from the providers above and thousands more. Your AI searches the catalog for free, picks a tool, and Frames pays the provider for that one call, in stablecoins, then checks that the data arrived. You pay the data cost plus 15% in credits, with one balance and no keys.

Every paid call is checked. Empty results, errors dressed up as success and calls that fail the delivery check aren't charged, and each request runs under a spending cap your plan sets. Every purchase gets a receipt: the tool, the price, the payment network and transaction, and whether it delivered. Your AI gets nine tools in total; the full list is on frames.ag/docs/mcp-tools.

How to add a data connector to Claude

The steps below use Frames' server address, and they're the same for any remote MCP server. This is the address you paste:

https://api.frames.ag/mcp

1. 1

   Go to Customize → Connectors, or open Frames' pre-filled add link.

2. 2

   Click + → Add custom connector, name it Frames, paste the server address and click Add.

3. 3

   Click Connect and sign in. In a chat, turn Frames on under + → Connectors.

Free plans allow one custom connector. On Team and Enterprise, an owner adds it first under Organization settings → Connectors. Once added, it also works in the desktop and mobile apps.

How to add a data connector to ChatGPT

1. 1

   At chatgpt.com, switch on Developer mode in Settings → Security and login.

2. 2

   At chatgpt.com/plugins, click +, name it Frames, choose OAuth and paste the server address.

3. 3

   Click Create and sign in, then type @Frames in a chat.

Web only, on paid plans. Menu names vary by account.

How to add a data connector to Cursor

1. 1

   Open the Add to Cursor link and confirm. Or add Frames by hand:

   {
     "mcpServers": {
       "frames": { "url": "https://api.frames.ag/mcp" }
     }
   }

2. 2

   Sign in when Cursor asks.

3. 3

   Ask the agent for live data.

Install links need Cursor 3.15.12 or later.

Claude Code, Codex and the rest

In Claude Code, one command, then sign in from the /mcp menu:

claude mcp add --transport http --scope user frames https://api.frames.ag/mcp

In Codex, the same idea:

codex mcp add frames --url https://api.frames.ag/mcp

VS Code with Copilot, OpenCode, Grok Bot and Hermes each have their own three steps on frames.ag/docs/mcp. Every setup above was checked against the app's own docs in September 2026. Apps move their menus now and then, so if a step doesn't match what you see, the app's help pages have the current path.

What to ask once it's connected

Ask the question you'd normally spend an afternoon on. Frames is for finding things out, so every prompt asks for information, and none of them buys, sends or books anything:

• "Use Frames to find which climate fintech startups raised a Series A in the last 90 days, with the round size and the lead investor."
• "Use Frames to compare today's price for the Vitamix E310 on Amazon, Walmart and Target."
• "Use Frames to pull what people on X are saying about our launch this week, full threads included."
• "Use Frames to find who leads partnerships at these ten companies, with a verified work email for each."

Two habits save time. If your AI answers without using the connector, ask it to use Frames by name. And in Claude, turn Frames on in each chat under + → Connectors. To see what a question cost, ask "How many Frames credits do I have left?", or open Settings → Usage in the dashboard.

What each route costs to run

A vendor's server costs that vendor's subscription, usually tens to hundreds of dollars a month, whether you query it or not. Multiply by the number of vendors you need.

Your own keys cost each vendor's API bill, plus the hours to build the server and keep it running as the APIs change.

Frames costs nothing to start. The Free plan gives you 3,000 credits a month with no card, and every plan reaches the whole catalog. 1 credit is $0.001, searching is free, and a request is capped at 1,000 credits on the Free plan, so a month of occasional questions fits inside the grant. Paid plans start at $19 a month for 10,000 credits; the full table is on the pricing page.

The missing piece was a way to pay

Claude, ChatGPT and Cursor already know how to ask for data while they answer; MCP gave them that. What they lacked was a way to pay for it. Connect one of the three routes and the free web stops being the ceiling on what your AI can tell you.

If a step doesn't work in your app, or a source you need isn't in the catalog, find us on Telegram.