Cloudflare status: hosting issues and outage reports

Problems in the last 24 hours

The graph below depicts the number of Cloudflare reports received over the last 24 hours by time of day. When the number of reports exceeds the baseline, represented by the red line, an outage is determined.

Most Reported Problems

The following are the most recent problems reported by Cloudflare users through our website.

• Cloud Services (30%)
• Domains (30%)
• Web Tools (17%)
• Hosting (13%)
• E-mail (9%)

Live Outage Map

The most recent Cloudflare outage reports came from the following cities:

City	Problem Type	Report Time
New York City	Cloud Services	1 day ago
Los Angeles	Cloud Services	3 days ago
Paris	Cloud Services	18 days ago
New York City	Hosting	21 days ago
Manchester	Domains	1 month ago
Angers	Cloud Services	2 months ago

Full Outage Map

Community Discussion

Tips? Frustrations? Share them here. Useful comments include a description of the problem, city and postal code.

Beware of "support numbers" or "recovery" accounts that might be posted below. Make sure to report and downvote those comments. Avoid posting your personal information.

Cloudflare Issues Reports

Latest outage, problems and issue reports in social media:

• 
  flavio (@flaviocopes) reported 4 minutes ago

  Cloudflare email sending service is very good

• 
  Ted (@tedbrine) reported 4 minutes ago

  I misconfigured a docker health check. In February 2025, I brought up a container. It was a turbo build cache. For some reason, I configured the health check to my portfolio. What is even more strange, is that I set the interval to five seconds. Now if this was heading into localhost, that probably isn’t the biggest problem. Heading into the internet kinda is. For the past 18 months, every five seconds, my server tried to send a request to my website. And it bypassed all Cloudflare protection because I told it to. The bigger problem is that if this failed, it would try again, and again, and again. 9.3 million requests. And now I know why my Vercel bill was big.

• 
  Jason Tally (@JasonTally) reported 8 minutes ago

  @EricLGoldstein @dok2001 You can already get a PlanetScale database through the Cloudflare dashboard and get billed for it through Cloudflare so I was counting that as one of the things that happened “another way”. From a customer perspective I don’t think a full acquisition would change much.

• 
  SPEKULATOR (@__spekulator__) reported 8 minutes ago

  @KennyJohnsonATX @Cloudflare oauth pre-registration is a real unlock for cloudflare's mcp server support. they can now hit slack and github without waiting on dcr support from them.

• 
  Astrid (@just_some_dev) reported 12 minutes ago

  cloudflare worker deploys seem down rip

• 
  Rheza Pahlevi (@rheza) reported 17 minutes ago

  @eastdakota @Cloudflare Can you guys please fix your support team first ?, I have been waiting for email support. My account and domain has been hacked and taken over by someone else.

• 
  dweewq (@eweqss1431) reported 18 minutes ago

  Agencies charge $8,000 to $12,000 for a marketing site, three weeks of calls, one invoice that makes you sit down. Claude Code builds the same site in an afternoon, but most people still get template output because they type "make it beautiful" and pray. Claude defaults to safe: Inter font, purple gradients, three feature cards. The ten thousand dollar look comes from constraints, not vibes. Screenshots beat adjectives. Three reference sites from your niche, with an explicit instruction not to copy the layout, give the model an actual quality bar. One prompt with five blocks, audience, the single action every page pushes toward, the references, the stack, and a banned list of clichés, gets a working first version in under ten minutes, about seventy percent there. The part that earns the price tag is the polish pass agencies bill forty percent for: typography, spacing, and motion, fixed in three separate messages instead of one, plus a mobile check at 375px since most traffic is a phone. Shipping costs nothing. Push to GitHub, connect Cloudflare Pages, deploy. The agency was always selling three weeks of process. The process was always one afternoon.

• 
  Xamayon (@SauceNAO) reported 19 minutes ago

  If you use Vultr Firewall Groups and have a rule restricted to Cloudflare sources, it's probably blocking all traffic right now... Either switch the rule to allow all or disconnect and reconnect the firewall group from your servers. Not thrilled that they knew about the issue and didn't bother to let their customers know.

• 
  Matt Parrott (@MatthewParrott) reported 21 minutes ago

  @KCDN19 It's dumb that I have to choose between Cloudflare models that are in the same warehouse but slow as **** for no good reason, Fireworks which is fast as **** but scams me on cache, and Communists on the opposite side of the world with no GPUs.

• 
  Val Vesa (@ValSocial) reported 23 minutes ago

  @Santosh74038967 Cloudflare Analytics measures server-side HTTP requests and network traffic at the edge, capturing all bots, crawlers, and direct hits. In contrast, tools like Google Analytics track client-side user behavior via JavaScript, counting only browsers that execute the script and attempting to filter out automated traffic.

• 
  BRAP (@BRAP_Podcast) reported 27 minutes ago

  @JCgamerX18 An outage doesn't mean there wasn't redundancy. AWS, Azure, Cloudflare, Xbox Live, Steam, PSN, and even banks all have backup systems, and they've all had outages. That's just the reality of running massive online services. If you think Sony isn't ready for a digital future, point to the actual architectural problem instead of assuming the outage proves it.

• 
  Rosed (@RosedInqually) reported 30 minutes ago

  Can X and Cloudflare stfu already I genuinely can't stand "Verify your age" just to click the same button over and over again and NEVER be able to actually verify, this **** sickens me.

• 
  Warya Wayne (@WaryaWayne) reported 31 minutes ago

  @zeeg Hey David, I'm a Sentry user. One task for Junior could be to maybe clean up the Sentry integration in Tanstack Start? Based on if we choose Nitro or Cloudflare it's dist or output and the scripts are wrong in package[.]json, there is a conflict my agents fix on deploy everytime

• 
  a_skin2 🔺 (@a_skin2) reported 32 minutes ago

  The Avax C-Chain API on @KoinlyOfficial seems to be malfunctioning (Cloudflare 502). Won't let me sync my txs since July 10. This is affecting my oldest wallet, the other ones seem to be working to some extent (missing ERC-20 txs and only tracking avax txs). Is anyone experiencing the same issue?

• 
  Kamran Wajdani (@kamranwajdani) reported 36 minutes ago

  @anilsoylu @Cloudflare that’s wild. if a bare test page can get flagged that fast, either the report flow is too loose or they’re using almost no signal before escalating it.

• 
  Huintellimance (@Huintellimance) reported 40 minutes ago

  Anthropic just deleted sessions from MCP — and it's the biggest protocol update since the spec launched. Here's what changed and why every AI developer should care: OLD MCP = a phone call. Your client and server ran an initialize handshake. The server returned a session ID. Every subsequent request carried that ID, and the server held your negotiated state in a live object inside one specific process. Think of a restaurant where only the waiter who took your order knows what you ordered. Works fine — until that waiter goes home. The problem: because state lived in one process, you couldn't spread requests across instances. Load balancers needed sticky sessions. Teams pushed state into shared storage. A single server restart dropped every open session. Autoscaling was essentially blocked. NEW MCP = an order slip. Anthropic killed the handshake and the Mcp-Session-Id header. Every request now carries its own protocol version, client identity, and capabilities in a _meta field. Any request can land on any instance behind a plain round-robin balancer. MCP servers are now ordinary HTTP services. They run on serverless. They run on edge. They survive restarts. But removing sessions broke three features, so each got rebuilt: Interactive tools → Servers used to push questions down a held-open stream. Now they return an "input_required" status and the client retries with the answer attached. Clean request/response. No persistent connection needed. Routing → Method and tool names moved into Mcp-Method and Mcp-Name headers. Gateways and rate limiters can route and meter without parsing JSON-RPC bodies. This is a massive win for API infrastructure teams. Caching → List responses now carry ttlMs and cacheScope. Clients cache tool catalogs instead of refetching on every reconnect. Less overhead, faster cold starts. Here's the part most people missed: State didn't disappear — it moved somewhere the model can actually see it. Instead of a session ID hidden in a header (invisible to the LLM), tools now return an explicit handle as an argument. The model can read it, thread it between tools, and recover from failed calls. A session ID in a header = the model has no idea it exists. A handle in the arguments = the model can reason about it. This is a fundamental shift in how agent state works. The protocol now treats the model as a first-class participant in state management, not a bystander. What this means in practice: Deploy MCP servers on Cloudflare Workers, Lambda, or any edge runtime — no session affinity required Horizontal scaling is finally trivial: round-robin, no shared store Server restarts don't kill active workflows Agent state becomes explicit and model-readable instead of hidden in infrastructure The MCP vs. CLI debate was already the wrong frame. This update makes it irrelevant. MCP servers are now as easy to deploy as any REST API, and agent state lives where it should — in the model's context, not in some ephemeral server process. If you're building agent infrastructure and haven't read the 2026-07-28 spec update, you're already behind. What's your current MCP deployment setup — and does stateless change your architecture plans? #MCP

• 
  bnistordev (@bnistordev) reported 41 minutes ago

  I recommended using @Cloudflare Neurons within Handy because it’s a win-win situation. The problem is that it keeps generating an endless number of errors of different kinds, and Cloudflare’s API returns a vast array of models. I spent at least 30 minutes trying to get everything working, but in the end, I couldn’t, so I switched to DeepSeek.

• 
  WIV Reports — Uncensored (@WIVReports) reported 45 minutes ago

  Any first-year network engineer knows the playbook: see a 429 with a CF-Ray ID, log into the Cloudflare dashboard, check the WAF/Rate Limiting logs for that ray ID, and adjust the rule. It’s literally a 2-minute fix. Instead, they insisted Cloudflare wasn't blocking anything, claimed DevTools were wrong, and issued a partial refund rather than taking 120 seconds to look at their edge logs. Pure laziness.

• 
  sam (@samgoodwin89) reported 45 minutes ago

  @jonas @teej_dv @alchemy_run We'll never produce a wrangler.json. We will just close the gaps and incorporate local development into our flywheel so that we have local dev for not just Cloudflare, but AWS, Azure, GCP, Railway, etc.

• 
  Jonas Templestein (@jonas) reported 47 minutes ago

  @dok2001 @jachands @mattzcarey I think even outside MCP oauth, managing scopes on access tokens on cloudflare is super confusing and tedious There seem to be two similar but different UIs that are each really fiddly and require advance knowledge of what scopes are called. There is no “manifest” support as far as I can tell where I could just paste a Jain blob or something to update scopes

• 
  nupjas (@nupjas) reported 49 minutes ago

  @rbayuokt @CloudflareHelp @Cloudflare seemed cf was down shortly but now working again for me

• 
  DrGlitchy ⚒️⚙️ (@DrGlitchyTV) reported 50 minutes ago

  @CarlosHenr36599 @realrusro You would have to train 1.5 million moderators before you could put them in the job, the would have to speak a litany of different languages and you would need even more people to watch **** groups and relay the new slang and abbreviations. You also need a network team to prevent those people from getting back on the platform (or you could hire Cloudflare for a custom integration).

• 
  Ben Badejo (@BenjaminBadejo) reported 51 minutes ago

  If you are debugging something that involves networking for hours and you don’t check for outages/maintenance from Cloudflare / AWS and so forth…you’re gonna have a bad time. Sometimes, the problem isn’t you. It’s them. 🙃

• 
  Anıl Soylu (@anilsoylu) reported 52 minutes ago

  @TheNicholasNick @Cloudflare I don't think it's at that point. I expect they'll fix it soon.

• 
  Onyx_Digital (@OnyxAudit) reported 54 minutes ago

  Spent 2 hours reverse engineering Cloudflare. Downloaded pages. Compared MD5s. Diffed HTML. Inspected __CF$cv$params. Compared cf-ray headers. Trying to figure out why Cloudflare was returning: Cloudflare Ray ID: 000000000000000 Your IP: 0.0.0.0 Root cause: $ protonvpn disconnect Site loaded immediately. I have become the Layer 8 problem. @ProtonVPN @theXSSrat I definitely need to join your classes.

• 
  Nic0le (@nicole_clash) reported 56 minutes ago

  @linguinelabs I have cloudflare but I also won vercel credits a while back so I never bothered 😭😭

• 
  GrainStats 🌾 (@GrainStats) reported 57 minutes ago

  @HTSCommodities @bfl_ai @NousResearch anything is possible right now i'm betting on agentic flows being monetized. there is a market for this, it will take time, but it will get done. that and IP rights being monetized (e.g. buy my chart, limited to 100 views) most of our traffic is bots these days anyway...maybe charge them a nickel per visit? surely cheaper than burning tokens for no reason no doubt in my mind that *everything* and i mean everything of value will be paywalled and locked down in the next few years (also thesis to long Cloudflare)

• 
  Benjamin Shelton (@zancarius) reported 59 minutes ago

  @kwrob24 Yeah... I have most of my blogs behind Cloudflare specifically for this reason, but for historical (read: lazy) reasons, this instance lives on a domain I use for a *lot* of personal projects and home-lab-esque sort of things, even if most of it has been migrate to cloud services. It's not really a huge problem, and it's not (yet) consuming a bunch of egress traffic. But, this isn't the first time I've had issues with AI-related crawlers. I may be forced to move this to another domain at some point.

• 
  Mat | HAMCORE (@TheHAMCORE) reported 1 hour ago

  @iPixelGalaxy I'd be interested to see where ClypDat goes. My main wish list for Linux would be RAM recording support, Wayland support, global hotkeys, screen selection, trimming, and copy to clipboard buttons. Currently Vice uses cloudflared and Cloudflare Tunnel to make links.

• 
  jamesb lawson (@JamesbL) reported 1 hour ago

  @SmorglesBord @AlecStapp he thought she was indian..lol Cloudflare caused a major global internet disruption on November 18, 2025, taking down popular services like X, ChatGPT, and Spotify. The outage stemmed from an internal software crash rather than a cyberattack. [1, 2, 3]