fly: a scraped list of free v2ray and Clash proxy nodes

Two subscription files, not a client and not a service

The repository is small enough that its whole inventory is visible at a glance. The top level holds a .github/ directory, a .gitignore, a LICENSE, the README, an addition/ directory, a clash file, a traffic.png image and a v2 file. The two data files are the product.

The clash file is a Clash subscription in the YAML form Clash clients consume, and the v2 file is the v2ray subscription in the form v2ray clients consume. Both are published as raw files on the main branch, and both are reachable through a mirror:

https://raw.githubusercontent.com/ts-sf/fly/main/v2
https://raw.githubusercontent.com/ts-sf/fly/main/clash

The mirror host, mirror.v2gh.com, is offered for the same two paths, so a client that cannot reach raw.githubusercontent.com can be pointed at the mirror instead.

What this repository is not matters as much as what it is. There is no client code, no configuration generator, no account, and no server operated by the project. The primary language field is not even set, because there is no program here. It is a published list plus a small amount of metadata about it.

The README also states the update policy in one line, that it updates at irregular intervals, which is an unusually honest description of a subscription that will change without warning.

The tool table shows which client each file is for

The README opens with a table mapping the supported clients by platform, and it is the practical starting point for anyone new to this.

For v2ray, Android uses v2rayNG and Windows uses v2rayN, both from the 2dust projects. For Clash, the Android column is marked with a dash and only Windows is listed, with Clash for Windows from the clashN releases.

That table is a small but honest description of the compatibility surface. The dash for Clash on Android is not an oversight in the formatting, it reflects that the project does not point Android users at a Clash client here. Anyone on Android is expected to be using v2rayNG and the v2 subscription.

The link targets are the release pages of the client projects rather than the clients themselves, so the assumption is that you already have, or are about to install, one of these clients and want the node list rather than the software. The addition/ directory is not explained in the README, and neither is the purpose of traffic.png beyond the heading it appears under, which is node speed test results.

The topics attached to the repository mention more protocols than the documentation describes, including shadowsocks, ss, ssr, trojan, v2ray, vmess, clash and vpn, while the README only documents the v2ray and Clash subscriptions.

A first use: import the subscription, then read the speed test

There is no installation step for this repository, because there is nothing to install. The workflow is to point a client at a subscription URL and let the client pull the list.

Concretely, you install v2rayNG on Android or v2rayN on Windows, or Clash for Windows if you want the Clash format, and you add the corresponding raw URL as a subscription in that client. The client then fetches the file and populates its node list. The mirror URLs serve the same content and are a fallback when GitHub raw is unreachable from your network.

Before you do that, look at the speed test image, which the README places under the heading for node speed test results, and a traffic.png file in the repository root. Since the list is refreshed at irregular intervals, the image is a snapshot of one moment rather than a standing measurement, and it will not reflect what the list looks like after the next update.

The sample nodes printed in the README illustrate the format rather than serving any practical purpose. Each line is a vmess:// URI carrying a base64 encoded JSON payload with the connection details, the port, an id, the security setting, the network type, the host, the path, the TLS setting and the host header. The sample includes entries for a domain fronting style host, for a plain ws network with no host, and for a node in another country, which is a fair illustration of the variety a scraped list contains.

Every node in the list is somebody else's server

The README states, in one line, that all the free nodes are crawled from the network and should not be used for illegal purposes. The second half of that sentence is the wrong half to focus on, because the first half is the operationally important one.

A scraped node list is a list of servers run by people you have not identified and cannot audit. Whoever runs the machine sees the connection metadata at minimum, and anything you send through a proxy that is not end to end encrypted is visible to that operator. The list contains no integrity checks, no owner identity, and no indication of how a given node came to be public in the first place, whether it was offered freely, misconfigured, or left running by someone who no longer wants it.

The consequence for a reader is that the value of the list is entirely in the convenience, and the risk is entirely in the trust. It is useful for a quick connection on an unfamiliar network, and it is the wrong place to carry anything that matters.

The project does make the free-nodes framing explicit, and the topic list makes the intent plain, so this is not a case of a project overstating what it is. It is a distribution list, and treating it as anything else is the mistake.

What the project does not document

The README is short, and it is worth being precise about what it leaves out, because the omissions are the same omissions you would want covered before pointing a client at it.

There is no documented collection method. How the crawler finds nodes, what sources it watches, and what filter it applies are not described, and the addition/ directory at the root is not explained. There is no stated node lifetime or expiry, and no statement about how a node that has stopped working is removed, which means a subscription can grow stale in either direction.

There is no privacy statement about what the project logs about the people using the list, and no policy on whether connecting to a node is reported anywhere. There is no mention of what happens to the v2 and clash files beyond the irregular update note, and no version or schema marker in them, so a client cannot tell an older format from a newer one.

The MIT licence covers the repository, which means the list files and the README are covered, but the licence says nothing about the servers behind the nodes, whose terms are unknown. The repository is not archived and the last push was on 2026-10-20, so the list is being refreshed as you read this, which is both the point of the project and the reason you cannot rely on any particular entry.

Editorial conclusion

Adopt fly if you want a ready made node list for a v2ray or Clash client and you are willing to treat every entry as untrusted, because the list is refreshed automatically from whatever the crawler finds and nothing in the project verifies where a node comes from or how long it will live. Do not adopt it for anything that carries credentials, private traffic or work data, since the repository is candid that the nodes are scraped from the network and offers no assurance about the operator behind any of them. Before you point a client at it, check what the current v2 and clash subscription files contain, confirm the mirror you use matches the raw file, and keep your own local copy of a known-good node so that a bad refresh does not leave you with nothing.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. README
4. ts-sf/fly on GitHub