Clamp (@clampdown19) header
Clamp (@clampdown19) profile picture

Founder of @AgentScoutHQ. Building a daily-tested directory of AI agents & MCP servers. Notes, stacks, real numbers.

Joined September 2023
Photo by Clamp (@clampdown19)Photo by Clamp (@clampdown19)Photo by Clamp (@clampdown19)Photo by Clamp (@clampdown19)Photo by Clamp (@clampdown19)Photo by Clamp (@clampdown19)Photo by Clamp (@clampdown19)Photo by Clamp (@clampdown19)Photo by Clamp (@clampdown19)Photo by Clamp (@clampdown19)
I'm building AgentScout, a free directory of 5,000+ AI agent tools and 3,000+ MCP servers. I post what I learn running it: setup gotchas, which MCP servers actually answer, and real traffic numbers. What are you building? agentscouthq.com/?ref=x @AgentScoutHQ @mdjunaidap #AIAgents #MCP #BuildInPublic
4
5
418
Switching branches with half-done work? Plain git stash leaves brand-new files behind, so they follow you to the other branch. git stash -u takes untracked files too. git stash pop brings it all back.
23
Stripe webhook signature check failing even though the secret is right? Your framework probably parsed the body first. Signature checks need the raw request body, byte for byte. In Express, put express.raw({ type: 'application/json' }) on the webhook route instead of express.json(), and register that route before the global JSON parser.
15
Port 3000 already in use and no idea by what? lsof -i :3000 It lists the process and its PID. Then kill <PID>. It's often an old dev server still running in a terminal tab you forgot about.
17
Your links show up on X and Slack with no image? Check og:image first. It has to be a full absolute URL, https and domain included. A relative path like /og.png doesn't work. For the big image card on X, also set twitter:card to summary_large_image.
19
Shipping a risky Cloudflare Worker change? You don't have to send it to all traffic at once. npx wrangler versions upload npx wrangler versions deploy Upload creates a new version without serving it. Deploy lets you split traffic between versions, say 10% new and 90% current. Watch the logs, then move it to 100%.
23
Staged the wrong file? git restore --staged <file> takes it out of the commit and leaves your edits alone. Drop --staged and it's a different command: git restore <file> throws the edits away, and there's no undo for that. Run git diff before the second one.
19
Spending time on priority and changefreq in your sitemap? Google ignores both. It does use lastmod, but only when it's consistently accurate. If every deploy stamps every URL with the build time, that signal stops meaning anything. Set lastmod to when the page content actually changed, and only list URLs you want indexed.
20
Retrying a Stripe API call after a timeout? Send an Idempotency-Key header, and reuse the same key on every retry. If the first request actually went through, Stripe returns the original result instead of creating a second PaymentIntent or charge. Build the key from the order ID, not a fresh random value per attempt. A new key on each retry defeats the whole point.
16
Writing a health check in a shell script? Let jq set the exit code. curl -s example.com/health | jq -e '.status == "ok"' -e makes jq exit 1 when the result is false or null, so the line works directly in an if, a CI step or a set -e script. No grepping JSON.
22
Need to look at a production database that isn't open to the internet? Don't open the port. Tunnel it: ssh -N -L 5433:localhost:5432 your-server Then point your local client at localhost:5433. The traffic goes through SSH, and the database stays private.
27
Fixed a page Google wouldn't index? In Search Console's URL Inspection, the first result is the version Google already has, not your fix. 1. Click Test live URL to see what Googlebot gets right now. 2. Check the rendered HTML and the indexability result. 3. Only then click Request indexing. Skip the live test and you can end up asking Google to recrawl a page that's still broken.
1
26
Want a Lighthouse score you can compare between deploys? Run it from the command line: npx lighthouse example.com --only-categories=performance --output=json --output-path=./lh.json Same settings every run, and the JSON has every metric, so you can diff today against last week instead of trusting a one-off run in DevTools.
22
Using Stripe Checkout? Pass your own user ID when you create the session, in the client_reference_id field. It comes back on the checkout.session.completed webhook, so you know exactly which account paid without matching on email addresses. Emails change; your IDs don't.
24
Shipped this week on AgentScout: our MCP servers page now hides servers we've found discontinued, behind a "Show discontinued servers" toggle in their own section. Dead projects were sitting next to live ones, and a list like that is only useful if what's on it still works. Right now 3 are hidden. agentscouthq.com/mcp/
1
27
Need to keep a PDF out of Google? There's no HTML head to put a meta robots tag in. Send it as a header instead: X-Robots-Tag: noindex It works on any file type: PDFs, images, plain text. And don't block the file in robots.txt too, or Google never fetches it and never sees the noindex.
1
32
Moved pages around? Check the whole redirect chain, not just the first hop. curl -sIL example.com/old | grep -iE "^(HTTP|location)" Each status line and Location header is one hop. If you see more than one 301 before the 200, point the old URL straight at the final one. Every extra hop is a wasted round trip.
1
31
Plain git stash leaves new files behind. Need a truly clean tree for a quick fix on another branch? git stash -u The -u includes untracked files. When you're back, git stash pop restores everything.
1
36
Lazy-loading every image on the page? Skip the big one at the top. loading="lazy" on your hero image delays the largest thing above the fold, which makes LCP worse. Load it normally and add fetchpriority="high". Keep lazy loading for images further down.
29
Yesterday we set 208 AgentScout pages to noindex on purpose: 97 short news stories and 111 alternatives pages that had never brought in a visit. They weren't wrong, just thin. Google judges quality partly across the whole site, so a pile of weak pages can hold back the good ones. Now we add about 3 new indexable pages a day, and each needs a real search behind it.
24
Moving a site to a new server? Test it before you touch DNS. curl --resolve example.com:443:203.0.113.10 example.com/ curl skips the DNS lookup and sends the request straight to the new IP, with the right hostname for TLS. If your page comes back, the switch is safe to make.
27