Ngrok
Day 48 after signup
[ngrok news] Your coding agent can drive ngrok now
The ngrok MCP server is live on every plan, even free. Plus: how vLLM keeps GPUs busy
Why it works
- Lead with one substantial product story, then connect it to practical examples and deeper reading.
- Use clear links throughout so readers can choose between trying the feature, watching a walkthrough, or reading more.
- Keep the roundup personable with brief first-person asides and a reply prompt at the end.
How it is built
- Personal greeting
- MCP server introduction
- Use case and learning links
- GPU article features
- Product update roundup
- Reply invitation and sign-off
Email copy
View in browser Hi, you@example.com! Your coding agent already has ngrok http <YOUR_FAV_PORT_HERE> on lock. Past that, it's often guessing on traffic rules and can't help you get a read on who's actually hitting your services. Room for improvement, eh? So, we built an MCP server to deal with that and more, and it's taking over this email. (Plus two special GPU-shaped things at the end.) Coding agent, meet ngrok. The ngrok MCP server lets Claude Code, Codex, Cursor, and other popular harnesses read and make changes to how you get stuff online with ngrok. Ask about your services, traffic, or usage in plain language, or say what you'd like to happen between your code and the internet and your coding agent will… just do it. It's also in Claude's connector directory, signs in through your browser, never sees your authtoken or an API key, and works on every plan. Connect your coding agent → Or point any MCP client straight at it: https://mcp.ngrok.com/mcp Rate-limiting bots for sport. I asked my coding agent to help me launch gpu.compare by removing the OAuth block, then capping every IP at 120 requests per minute. It checked that no genuine visitor would hit the cap, read the docs, crafted the rule, validated it, applied it, and made sure it worked. All from one sentence! I wrote about that, blocking vulnerability scanners, why your coding agent still can't do more than you can, and what's changing about how you build with ngrok. Read the whole story → Get the house tour from Ryan. Watch Ryan walk through the ngrok MCP server end to end by deploying a secure Yodaspeak app, all in under four minutes. Watch Ryan explain → Ready to just ask for what you want to do with ngrok? Start here: "What's online right now?" "Are any of my endpoints on fire? Show me 500s by hour for the last day." "I got ERR_NGROK_8012 when I opened my ngrok URL. What's wrong?" "Which of my API keys are more than a year old, and who owns them?" Your coding agent answers straight from your own account. Easy peasy. And that's about enough MCP for one email. Annie, meanwhile, spent the month learning exactly how vLLM serves open-weight models at scale. vLLM from the forward pass to speculative decoding. Annie's deep dive shows you all the tricks vLLM, the open-source inference and serving engine, uses to keep GPUs maxxed out: continuous batching, prefix caching, chunked prefill, and more. Thanks to Annie, I finally know what a mixture of experts really is, and that, apparently, hot dogs are tacos. Make your GPUs earn their keep → But can you actually find a GPU to run vLLM on? Turns out that part might be harder than you'd think or like. So I built gpu.compare: every hour, it checks popular GPU providers for prices and whether you can actually rent them. It's free and you don't have to sign up… just the way more things in life should be. Find a GPU you can actually rent → We also: Added Claude Fable 5.2 (20% cheaper than 5.1) and Gemini 4 Carbon (at its launch price through Nov 30) to the AI Gateway, plus Kimi K3.1 Pro, GPT-6.1 Terra, Hy4, and Gemma 4. Taught the AI Gateway to talk and sing: text-to-speech from six providers, and Google's Lyria writing full songs, like our sea shanty about a webhook that never gets its 200 OK. Brought in Liquid AI's LFM3 models and 160+ more from LLM API. Gave the AI Gateway a usage page that splits each request's cost into text, images, tools, and thinking. Wrote a guide for moving from ingress-nginx to the ngrok Kubernetes Operator. See the full changelog → What's the first thing you'd ask your coding agent to do with ngrok? Hit reply and tell me. I read every one, and the best ones will end up as prompts in our docs. ~ Joel at ngrok Your trust in us matters. Learn more about security at ngrok. ngrok Inc. 2261 Market Street STE 71467 San Francisco California 94114 United States You received this email because you are subscribed to Monthly Newsletter from ngrok Inc. . Update your email preference