Connect your X account to Claude Code so it can read your own bookmarks, check whether what you saved is actually true, and go deeper on the ones worth it. This is how Munim works.
Why bother
Saving a post is easy. Working out whether it is real is not. A lot of what circulates about building with these tools is exaggerated, some of it is auto-generated filler, and a little of it is a straightforward scam selling a course.
Once Claude can see what you saved, you can hand it the whole pile and ask it to do the work you would not do by hand: check each claim against the actual source, tell you which ones are hollow, dig into what the person is really doing, and find whoever is doing it better. It turns a saved list into something that answers back.
What it costs
There is no subscription and no minimum. X charges only for what you read. Their words: "No contracts, subscriptions, or minimum spend. Start and stop anytime."
What
Price
Reading a post
$0.005 each
Reading your own posts and saves
$0.001 each
Asking only "how much is being said about this" and getting a number back, no posts
$0.005 per question
So reading everything you have ever bookmarked is somewhere between about 10 cents and a dollar, depending which of those rates your saves fall under. Munim's whole list came to 99 posts, which is one go. Put a small amount of credit on it, ten dollars rather than a hundred, and see what a real sweep actually costs you before adding more.
Do this before anything else. In the console, set a maximum spend per billing cycle and leave auto-recharge off. Both are settings, both take a moment, and together they mean the worst case is that things stop working rather than that you get a bill. We run ours exactly this way.
Setting it up
Go to console.x.com and sign in with your normal X account. Accept the developer agreement when it asks.
Create an app. It wants a name, a description and what you plan to use it for. Reading your own bookmarks is a perfectly ordinary answer.
Save the credentials somewhere safe, immediately. They are shown once and never again. A password manager is the right home. Not a chat, not a note, not a message to anyone including Munim.
Add a little credit, then set the spending cap and turn auto-recharge off. In that order.
Turn on the permission that lets it read your bookmarks. This is the fiddly step, see below.
Tell Claude where the credentials live and let it do the rest.
The one genuinely fiddly step
Most things you can do with this need only a simple key. Your bookmarks are different, because they are private to you, so X wants proof that you personally said yes rather than just proof that the app exists.
In practice that means turning on the sign-in-with-X option in your app's settings and allowing the permissions called bookmark.read, tweet.read, users.read and offline.access. That last one is the difference between it working today and it working next month, so do not leave it out.
Expect this bit to take a couple of attempts, and let Claude drive it. It took us more than one go too, and there is nothing you can break by getting it wrong: the failure is that it does not connect yet.
Then hand it over
Once the credentials exist, paste this into Claude Code and work through it together.
I have an X developer account with credentials saved. Help me connect Claude Code so you can read my own bookmarks. Set up the sign-in permission properly, including offline access so the connection keeps working. Keep the credentials in a file only I can read, never in a message and never anywhere it could get committed. Then read my most recent 25 bookmarks and tell me what you found.
Copied
And once it works, this is the thing actually worth having.
Take my recent X bookmarks. For each one: read the real source, not just the post. Tell me whether the claim holds up, whether it is genuinely useful or just filler written to get attention, and whether anything about it looks like a scam or a course being sold. Then go and find who is doing the same thing better, and what they do differently. Be blunt about the ones that are not worth my time.
Judge them by these rules, not by popularity. A post saved far more often than it was liked is usually worth reading, because saving means someone wanted it again later. A quiet post can still be the best one, so use the numbers to decide reading order and never to decide what is true. Ignore enthusiasm and tell me what the person actually did and what broke while they did it. Treat a post that mentions no failure at all as probably selling something. For each one, tell me who the author is and what they gain if I believe them. Never invent an engagement number: if you cannot see it, say so.
Copied
How to tell a good post from a bad one
These are the rules Munim's own setup uses when it reads X, learned over a year of monthly sweeps. They are worth knowing yourself, not just handing to Claude, because they are mostly about resisting the obvious signal.
Saves beat likes. If many more people bookmarked something than liked it, it is usually worth your time. Saving means someone wanted it again later. Liking costs nothing and is mostly politeness.
A quiet post can be the best one. The single most useful idea this method ever turned up came from a post almost nobody had engaged with. Use the numbers to decide what to read first, never to decide what is true.
Ignore enthusiasm. "This changes everything" is not a finding. What did the person actually do, and what broke while they did it.
A post with no failure in it is usually selling you something. Anyone who has really built the thing will mention what went wrong first. Look for that, and look for whoever is arguing the opposite.
Always ask who is saying it and what they gain if you believe them. A name you do not recognise is not a reason to dismiss it, but it is a reason to check.
How not to waste money on it
We have overspent on this three times, so these are earned rather than theoretical.
How much you ask for matters. How often you ask does not. Reading the same post twice in a day is free. Reading twice as many different posts costs twice as much. Watching three busy accounts closely cost us about $10 in three days, which was nearly three quarters of that month's spend.
Ask for 25 things, not 100. One research session cost $1.91 from eight questions purely because the number was set high out of habit.
To find out whether a topic is loud right now, ask for the number rather than the posts. Half a cent instead of a dollar.
Never work out a budget from a day that has not finished. Ours came out four times too low that way. Assume the busiest day is at least double the first full one you measure.
Where this could be wrong
The two prices above do not quite agree with our own bill. X's price list says half a cent a post today. Our own account, measured across a month last summer, worked out at about a cent. So treat the figures here as the right order of magnitude and check your own console after the first real sweep. That is the number that counts.
The setup steps are from X's own instructions, not from watching someone do it fresh today. Ours was set up a while ago and their console changes. If a screen does not match, trust the screen.
This is how Munim happens to do it, which is not the same as it being the best way. If Claude finds a cleaner route while you are setting it up, take that one.