On sites like X and Reddit, the posts and replies are often visible only after you sign in. A research run that only scans the public web gets a login wall, or a short preview. The part you need to check is behind the wall.
The ways people connect those sites today are awkward to actually use.
The research tool fetches from its own servers. The site wants a login, the server does not have your session, and what opens is the login page. That material is missing from the run, or the model writes on from the title.
You hand the tool a cookie or an account and let it sign in for you. One research task is not a reason to give away an X or Reddit session, and afterwards you do not know where else that session was used.
You build an MCP server or a connector, handle auth, and keep it up when the interface changes. Most teams will not do that just to read through a set of posts. Those sources stay outside the research.
What we did
Those three run into the same wall. The page you can see is signed in inside the browser you are already using. The research tool does not have that session.
A server fetch is missing exactly that session. Handing over the account trades it for a risk one research task should not carry. Building an integration for each site puts the cost on the person who just wanted to read the page. In this browser, you can already open it. What was missing was a way to bring that step into the research.
That is why we built the browser connector. It sits in the desktop browser you are using. When a run needs a page, it uses the session already signed in there, opens what you can see at that moment, and hands the reading back to this run. You do not enter the account or the password into SpanSearch, and you do not hand over cookies or request headers. How you sign in to X or Reddit stays your business. The connector opens the page only when the run needs to read it.
A site that requires a login can then enter the research, without an MCP built first and without giving the session away. Built-in sources such as X and Reddit are read this way. If you are not signed in in this browser, the source shows that it cannot connect. Open the site, sign in, come back to SpanSearch, and try again.
A catalog also cannot cover the site each person actually uses. Asking them to build an integration for it is the same dead end as before. A site that is not in the catalog can be added from Data sources. Paste the URL, check the preview is the right site, and write what it can provide and when it should be used. Later research treats it like a built-in source: when the run needs to read, it still opens the page with the session in your browser.
How to use sources
Once a source is connected, you can name which ones this message should use before you send it. One is enough. Several is fine. If you already know where the answer should come from, pick them yourself.
You can also leave them unselected. When you do not name any, SpanSearch matches among the sources that currently connect, based on the question.
How to install the extension
The reading above happens only after the connector is in the desktop browser. Chrome, Edge, Brave, and Opera all work. Install the SpanSearch Browser Connector from there. After it is on, confirm the extension is enabled, then refresh SpanSearch.
If a source will not connect, refresh the page and try again. If it still will not, contact us.