mistgate Docs
EN
300

In the admin: Nodes

WARP

How to send a profile's traffic out through Cloudflare WARP, how to get an account for a node, what fail-closed means, and how to read and fix the WARP link.

On this page

By default a profile's traffic leaves a node from the node's own address. With WARP it leaves through Cloudflare's network instead: sites see a Cloudflare address, not your hoster's. A node has at most one WARP account, and every profile with the WARP exit on that node goes through it. Hysteria2, VLESS and AmneziaWG profiles can use it.

How it fits together#

Each profile on a node has its own exit. A profile with the Direct exit leaves from the node's address; a profile with the WARP exit leaves through the node's one WARP account, so sites see a Cloudflare address. Without working WARP the WARP-exit profiles carry no traffic and never fall back to the node's address. Apps Happ, kl!ck, AmneziaVPN Node de1.example.com Hysteria2 Exit: Direct Hysteria2, WARP copy Exit: WARP AmneziaWG Exit: WARP Direct node's IP WARP one account sites see the node's IP sites see a Cloudflare IP Fail-closed Without working WARP, the WARP-exit profiles carry no traffic: they never fall back to the node's own address.
  • The exit is a profile setting. Exit (in Advanced for Hysteria2 and VLESS, in Basics for AmneziaWG) is Direct or WARP.
  • The account belongs to a node. It is set up on the WARP card of the node: node → Exit, the first card. The WARP chip in the node's header leads there too.
  • A profile with the WARP exit needs an account on every node it runs on. The editor lists the nodes of the profile that have no working WARP, each a link to its WARP card.
  • MCP tools. warp_status reads the account and the node's last probes. With an Admin token, warp_restart_plan / _apply restart the tunnel; warp_reregister_plan / _apply replace the account after the owner approves the plan and sees the terms link.

To offer people both exits, make a twin of the profile with the other exit ("WARP copy"): see Profiles.

Fail-closed#

A profile with the WARP exit never leaves directly. If WARP is not there, the profile does not carry traffic, instead of silently showing people's traffic from the node's own address:

Situation What happens to WARP-exit profiles on the node
The node has no WARP account They do not start ("There is no working WARP on the node").
WARP is paused They stop passing traffic until it is resumed.
The tunnel is down They stop passing traffic until it comes back; if it is only slow, they work slowly.

The add-profile dialog warns before the click when the node has no WARP, has it paused, or its tunnel is down. The doctor's WARP exit check and the client-eye checks show the same from the other side: a node whose direct profile works while its WARP profile fails gets an alert that names the WARP exit. See Health.

Getting an account#

The card of a node without an account says Not set up and offers two ways. Both need the owner and a fresh confirmation of the sign-in (step-up), and both need an agent that supports WARP; an older agent gets a link to Updates.

Enable WARP (register)#

Enable WARP registers a new Cloudflare WARP account for this node and stores it encrypted. The window says what the click does:

  • it is an anonymous device account: no e-mail, no password;
  • the request leaves from the panel's address, not from the node;
  • it accepts the Cloudflare WARP terms on your behalf, for this one account. The window links the terms, and Enable WARP stays off until you tick "I have read the Cloudflare WARP terms and accept them for this account".

It takes a few seconds; then the node brings the tunnel up. If Cloudflare refuses new accounts from the panel's address ("Cloudflare refuses new accounts from the panel's address right now"), the window offers Import wgcf profile.

Import a wgcf profile#

Use this when registration from the panel does not work. Make the files with the wgcf tool on a computer where it works, then paste them or open them with Open file:

File Needed What it gives
wgcf-profile.conf yes The private key and the peer. It must have [Interface] with a PrivateKey and a [Peer].
wgcf-account.toml no The access token. Without it, Read from Cloudflare again does not work and Delete account removes the account only here, not at Cloudflare.

The WARP card#

Next to the title the card shows where the account came from (Registered here or Imported), the account type (free or plus) and the Cloudflare data centre of the last probe.

States#

State Meaning
Not set up No account on this node.
Waiting for the node The account exists, the node has not reported yet or is offline.
No backend The host has neither a WireGuard kernel module nor /dev/net/tun.
Starting The tunnel is coming up. "Starting: check failing" when the checks keep failing.
Working Handshake and probes pass. "Working: last check failed" when the latest check did not.
Not working The handshake or the probes fail.
Paused You paused it.
  • Handshake: when the tunnel last completed a handshake with Cloudflare.
  • Endpoint: the Cloudflare address and port in use.
  • Traffic: bytes in and out through the tunnel.
  • Cloudflare and Other site: two probes through the tunnel, one to Cloudflare and one to a site outside it. Each reads ok with its time, slow (over 2 seconds), or failed. A probe times out after 20 seconds; HTTP failures show their status.
  • "Checked N s ago". The node checks every 30 seconds, every 5 while the tunnel is starting. A run of failures shows as "3 failed checks in a row".
  • Details: the WireGuard backend, Cloudflare's own flag (warp=on, warp=plus or warp=off) and whether the account's reserved bytes are stamped on packets.

When a check fails, the card says what happened and what it means, in words: a probe failed, traffic does not go through WARP (warp=off), the handshake is stale or missing, the interface is down, or the host has no WireGuard backend.

Recovering by itself#

The health panel's Doctor offers Reconnect WARP when it confirms that the tunnel is down. Review and confirm the plan first. It rebuilds the current tunnel and reapplies its routes without changing the account. Profiles using WARP may pause briefly. A slow but successful probe does not trigger a reconnect.

While the tunnel is down, the node climbs a ladder, one step per failed check:

  1. it sets the routes, rules and device up again;
  2. it tries the other Cloudflare addresses and ports: ports 2408, 500, 1701 and 4500, the IPv4 endpoint first, then the IPv6 one if the account has it;
  3. it asks the panel to read the account from Cloudflare again (this needs the access token and creates nothing);
  4. it gives up and the card says Needs your attention.

After the last step it rests for 10 minutes, still trying the next address after each failed check, and then starts over. The card shows the step it is on ("Trying to recover: switched the address and port to …").

Buttons#

Button When What it does
Restart WARP The account is on and not revoked; the main button when the tunnel is down Pauses and resumes in one step: the node tears the tunnel down and brings it up again with the ladder from the start. If the node did not confirm the pause in time, the resume goes out anyway: "WARP is on again, but the node did not confirm the pause: look again in a minute".
Register again Always; the main button when Cloudflare no longer knows the account (revoked) Registers a new account and puts it in the old one's place in one step; the old device is deleted at Cloudflare. Same terms window as Enable WARP. Use it, not Delete account followed by a new registration, to get a fresh device for a slow or flaky exit: the node keeps its WARP until the new account arrives, so nothing stops and no alert opens.
Read from Cloudflare again The account has an access token Re-reads the endpoint, addresses and client id from Cloudflare. Creates nothing.
Pause / Resume Always Pause tears the tunnel down. The window names the profiles that stop and how many connections go through them now. A paused account is resumed, not restarted.
Delete account Always Type the node name to confirm; needs a step-up. The node loses WARP; WARP-exit profiles stop. The device is deleted at Cloudflare too when a token is stored. You can register or import another account afterwards.

Some problems need you, not a button. The card names them:

  • the routing table WARP uses (51820) holds another program's routes, often wg-quick: free the table or stop the other tunnel;
  • another program uses the routing-rule priorities WARP needs (90 and 110);
  • a network interface named mgwarp already exists and is not WARP's;
  • the node could not apply the account: the reason is in the node's events.

IPv4 and IPv6#

  • The account stores Cloudflare's endpoint as IP addresses, never as a name. The node tries the IPv4 endpoint on every port first, then the IPv6 one, if the account has it.
  • New accounts often have no IPv6 address inside the tunnel. Then the WARP exit dials IPv4 only, and IPv6-only destinations are not reachable through WARP.
  • A node without IPv6 is fine for WARP: the doctor notes that WARP goes through its IPv4 address.
  • Names are resolved with the node's DNS resolvers for this node, the same as for the direct exit. See DNS.

What the host needs#

A WireGuard kernel module, or /dev/net/tun for the userspace one. If Health → Doctor reports an outdated unit, apply Refresh the service there; the agent restarts and connections can drop for a few seconds. For an agent too old to offer that fix, run mistgate-node install once with the current binary.

Registration settings#

Two things about registration are not in the admin, only in the API (WarpService.GetWarpRegistrationParams and UpdateWarpRegistrationParams, owner only):

  • the values the registration request is built from (the API version, the user agent, the client version and the TLS fingerprint). Cloudflare changes what it accepts from time to time. An empty value falls back to the built-in one, and the read call returns the built-in values next to yours;
  • automatic re-registration, off by default. When it is on and Cloudflare revokes a registered account, the panel registers a new one by itself, at most once an hour per node. Turning it on accepts the Cloudflare terms for those future accounts too.

See API.

Note

Each node uses one free consumer WARP account, and the traffic of everyone on that node goes through it. Cloudflare may limit or suspend such an account; then the WARP-exit profiles stay down until you register or import another one. Do not rely on WARP for anything that must not fail.

Edit this page on GitHub