Shadowrocket Protocol Selection and Settings Guide
Choose a protocol your server supports, then check authentication, transport and rules.
Use this guide to look up protocol basics, field meanings and connection limitations. New users should start with the quick-start tutorial to import a configuration, connect and verify access, then return here to troubleshoot specific issues. The examples assume you already have your own subscription or server details and do not include usable connection credentials.
Shadowrocket is available across platforms. The iPhone / iPad version costs $2.99 as a one-time App Store purchase; versions for Mac (Apple silicon and Intel), Windows, Android and Linux are free. See the downloads for all platforms. Check the App Store listing for iPhone / iPad system requirements.
Protocol, transport and routing rules
Separate the roles of these three layers to avoid mistaking a connection failure for a protocol performance issue.
A protocol name is not a performance rating
A protocol defines how the client and server format requests, verify identity and transfer data. It does not determine server distance, outbound bandwidth, network congestion or how quickly a destination responds. The same protocol can perform very differently across servers, and results can change when the underlying network changes. So don’t start by looking for the “fastest” protocol name. First confirm what your existing server supports, then compare compatible options for ease of maintenance and real-world performance.
Think of a connection in three layers: protocol, transport and security. VMess or VLESS describes the proxy session; WebSocket and similar methods provide the transport; TLS encrypts and authenticates the relevant connection. These layers can be combined, but they are not interchangeable. Two entries with the same name in a subscription are not equivalent test conditions if their ports, transports, paths or certificate names differ. Changing the type manually does not change how the server listens, and can lead to timeouts or handshake errors.
Check the full settings in Add Server
When adding an existing server from Home, choose the type specified in your server details, then enter the address and port. Check the password or UUID, encryption method, transport and any additional fields. A blank field in the app doesn’t mean the server has no requirement, and a subscription display name is not a connection address. For imported subscriptions, keep the original entry and test on a duplicate so you can always restore the original settings.
The server address and certificate name may differ: the address determines where the connection goes, while the name helps verify the remote identity. A port is just a listening endpoint; it does not reveal the protocol. The same port number can serve different protocols, and the same protocol can use different ports. If information is missing, ask the configuration provider what the field means instead of guessing from a familiar port number or the app’s default. Listing the fields before troubleshooting is more useful than repeatedly toggling settings.
Global Routing does not change the server protocol
Global Routing options Config, Proxy and Direct determine how requests are handled. Config routes requests according to the active configuration’s rules; Proxy sends applicable traffic through the selected proxy; Direct connects without a proxy. These options decide whether and where requests are routed—they do not convert Shadowsocks into Trojan. If a domain works under Direct, that doesn’t confirm the proxy settings are correct. If it works under Proxy but fails under Config, check the rules and selected policy.
| Layer | What to check | Common misdiagnosis |
|---|---|---|
| Server connection | Address, port, protocol, authentication, TLS and transport | Assuming an open port means authentication succeeded |
| Request routing | Global Routing, rule order and policy target | Treating a DIRECT request as a proxy test |
| App access | DNS, destination address, TCP or UDP requirements | Blaming every connection for one website’s error |
Tests should also distinguish short connections, sustained downloads and real-time interaction. Short connections are more sensitive to the initial handshake and DNS lookup; sustained transfers depend more on available bandwidth and congestion control; real-time apps are affected by jitter and packet loss. A single Connectivity Test result is no substitute for testing your actual use case. Repeat the same task on the same network and note whether failure occurs before connection, when data transfer starts or after some time.
Shadowsocks: encryption method, password and UDP limits
Settings are relatively simple, but check the encryption method and datagram support together.
Understand the overhead behind encrypted proxy connections
Shadowsocks is designed to carry proxy requests over an encrypted connection with a relatively simple data path. It typically avoids extra HTTP request encapsulation, which can make troubleshooting more direct when the server offers a native connection. But fewer layers do not guarantee higher speed in every situation. Server processing power, implementation, network quality and the selected method all affect performance, especially when switching mobile networks or transferring data for a long time. Don’t predict performance from the protocol name alone.
Distinguish older stream ciphers from AEAD methods. AEAD provides both confidentiality and integrity protection, allowing the recipient to detect modified data. Methods differ in key length, message format and implementation requirements; similar names do not make them interchangeable. Match the client’s method to the one configured on the server. If the imported details specify a newer protocol variant, confirm that both client and server support it rather than copying only the password.
What Method, Password and address each do
The address and port locate the listening service; Method determines how both sides process encrypted data; Password helps establish the correct encryption context. An incorrect password may still allow a network connection to the port, but the server cannot process the data correctly. You may see timeouts, a closed connection or a successful test without an actual page loading. Changing DNS repeatedly won’t fix incorrect authentication or a mismatched method. Check each field against the original server details.
When entering settings manually, watch for leading or trailing spaces, capitalization and keyboard substitutions in the password. Don’t use an encoded segment from a share link as the password, or copy symbols from a display name into the port field. If the imported entry includes additional transport settings, confirm that the server specifically requires them. Extra layers are not universal enhancements; enabling one arbitrarily may send data the server cannot recognize. A complete subscription usually preserves more required details than manually copying selected fields, but a successful import still needs to be verified by accessing a real destination.
TCP working does not mean UDP is working
Web browsing often reveals whether the TCP path works, while real-time audio and video, some name lookups and certain apps may use UDP. Shadowsocks UDP support depends on the server forwarding datagrams, and the network path must allow them to travel both ways. A related client setting does not mean the remote server has UDP enabled. If web pages load but a real-time feature fails, check whether the app needs UDP, then verify the server configuration and network conditions instead of blaming the password for every issue.
UDP datagrams also incur encryption and encapsulation overhead. If a packet is already close to the path’s size limit, encapsulation can cause fragmentation or dropped packets. Small requests working while larger datagrams fail is a useful diagnostic clue. Don’t immediately change network settings across the whole device. With the same server, compare another network or another explicitly supported datagram path to identify whether the cause is protocol settings, server forwarding or the access network. Repeatable conditions are more useful than a one-off success.
How to assess speed and battery use on iOS
Energy use depends on processor capabilities, implementation and traffic volume; no encryption method is inherently a fixed battery-saving option. With intermittent browsing, waking the wireless radio and establishing connections may use more power than encryption. During sustained transfers, throughput and network retransmissions matter more. For a meaningful comparison, keep screen brightness, app workload and signal conditions similar, and record how long the task takes to complete—not just the peak speed.
If you already have a native Shadowsocks server with complete settings for ordinary app use, start with it as a baseline. If the server explicitly requires another transport, follow its instructions rather than removing required fields to simplify the setup. Before migrating or deleting entries, keep a restorable copy of the original configuration and check whether subscription updates will overwrite local changes. See managing multiple subscriptions for grouping subscriptions and handling duplicate entries.
VMess and VLESS: understand protocol and transport combinations
The same UUID does not make the two protocols interchangeable; transport settings must also match.
The two protocols handle the security layer differently
VMess combines user authentication with its own protocol-level data protection. The server must identify the user and use a compatible protocol implementation. VLESS keeps the protocol layer lighter and relies on the selected security transport for data protection; VLESS itself does not encrypt payloads. So when you see VLESS, check whether TLS or another server-specified security layer is required. Don’t treat VMess and VLESS as the same protocol with a “speed” toggle.
Older VMess configurations may include legacy authentication fields, and requirements differ between modern and older deployments. If imported details include such a field, follow the server’s actual configuration and the current client’s support rather than changing it to a fixed value found online. Keep the device clock accurate, as authentication may depend on a time window. If the time is significantly off, restore automatic time before reconnecting instead of repeatedly generating new user IDs.
Keep UUID, transport and path settings together
A UUID identifies a user; it is not a server address or an arbitrary note. Copy the complete value and make sure no characters are missing. The transport determines how both sides establish the underlying connection. If the server expects WebSocket, a client sending a plain protocol stream cannot negotiate successfully, and vice versa. Confirming only the UUID and port isn’t enough to verify a complete configuration, especially when a reverse proxy sits in front of the server.
WebSocket commonly uses Path and Host. Path matches the relevant server route; Host may be used to select a virtual host at the entry point. These are different from the TLS Server Name, even if a deployment happens to use the same value for some of them. Preserve the original path, including capitalization, trailing slash and query string where present; “cleaning up” the string can change how it matches.
| Field | Purpose | What to check |
|---|---|---|
| UUID | Identify the protocol user | Complete value and whether the user is still active |
| Transport | Choose the transport | Must match the server’s listening configuration |
| Path / Host | Specify the HTTP transport entry point | Keep the original value; don’t confuse it with the certificate name |
| Server Name | Used in TLS name negotiation and verification | Must match the server certificate and entry requirements |
Benefits and costs of extra encapsulation
WebSocket and similar transports are generally used to fit an existing server entry point or reverse proxy setup, not to automatically increase speed. Additional handshakes, request headers and framing add overhead, but the real impact depends on connection reuse and task size. Apps making many short requests may be more affected by the initial connection cost, while sustained transfers may be limited mainly by the server’s outbound capacity. Compare complete configurations, not just the names VMess and VLESS.
Some configurations may also include dedicated flow-control or negotiation fields. They matter only with a matching protocol, security layer and server implementation; they are not generic speed-boost settings. Don’t add a field if it wasn’t provided in the original subscription. If the client doesn’t recognize a combination, first check its support, then ask the configuration provider for compatible connection details if needed. Don’t delete unfamiliar fields and overwrite the original entry.
Making a meaningful comparison on iOS
Where possible, use the same server location, similar exit conditions and the same destination. Compare the first load, repeated visits and recovery after backgrounding. Extra encapsulation may not cause a noticeable difference, but frequent reconnects on a weak signal can make connection setup more costly. Battery comparisons should cover the actual task and account for differences such as file size and screen-on time; otherwise, task differences can be mistaken for protocol energy use.
If two entries use the same address but only one works, check UUID, transport, security layer, Path, Host and Server Name in that order. This is usually easier than resetting every field at once. If both connect but only certain destinations fail, investigate rules and DNS next. Performance comparisons are valid only after confirming that requests actually use the selected proxy. A request routed to DIRECT under Config cannot tell you which protocol performs better.
Trojan and TLS: the connection address is not the certificate name
Separate reachability, TLS identity verification and protocol authentication to pinpoint failures.
Why Trojan relies on TLS
A typical Trojan setup uses TLS to protect the client-server connection, then handles proxy authentication and requests within it. Because TLS is part of the design, the certificate, name and handshake are essential—not optional extras. Reaching the server port only shows that the network path allows a connection. It does not confirm that the certificate is valid, the password is correct or the destination request was forwarded. Test each stage separately when troubleshooting.
TLS provides both encryption and remote identity verification. Encryption prevents observers along the way from directly reading plaintext; certificate verification helps the client confirm that it has reached the expected service. Neither replaces the other. An encrypted connection to an entry point is no reason to ignore a name mismatch or certificate chain error. Keeping verification enabled helps prevent a wrong entry point from being mistaken for the correct server.
How Address, Server Name and certificates relate
Address specifies the connection destination, which may be a domain or an address supplied by the server. Server Name is generally used for TLS name negotiation and should correspond to the expected certificate identity. If you connect by address but the certificate is issued to a domain, enter the corresponding name as specified in your server details rather than copying the address into every field. Some entry points also route by name, so an incorrect name may lead to a different certificate or site.
Certificate verification usually checks validity dates, a trusted issuer chain and a matching name. An incorrect device clock can affect validity checks, and a server missing an intermediate certificate may fail on some devices. Check the device time, address and name first, then verify the server’s certificate chain. Don’t routinely disable verification to “fix” a connection: that removes identity checks and hides server configuration problems, making future migrations and troubleshooting harder.
Password and application-level negotiation still need to match
Completing TLS does not mean Trojan authentication has succeeded. The Password must match the information provided for the server; avoid adding spaces or line breaks when copying it. Preserve any required transport or entry settings as well. A matching password does not mean you can reuse the address and port from another service. Different listening endpoints may use different protocols or authentication databases, and mixing fields can result in an apparently connected session with no working access.
If the app offers options such as ALPN, they are for application-layer protocol negotiation, not arbitrary performance labels. Unless the server explicitly requires a value, keeping the original settings or default negotiation is safer. The server, front-end entry point and client must have a compatible option in common; forcing an unsupported value can break the handshake. When troubleshooting, note whether the error occurs before TLS is established or after authentication to narrow down what to check.
Handshakes, connection reuse and battery use on mobile
Establishing TLS requires communication and processing. When a connection can be reused, the cost is spread across later requests. If the network changes frequently or connections keep dropping, setup costs recur. To assess Trojan on iOS, look beyond a single cold-start delay: observe repeated access on a stable network and recovery when switching from Wi-Fi to cellular. A long recovery may be caused by the network, entry point or DNS—not necessarily the protocol.
During sustained transfers, power use also depends on throughput, signal strength and server response. Finishing the same task faster may reduce the time the wireless radio stays active, but a higher peak speed does not automatically mean lower battery use. Compare total time for the same task and performance over a sustained period. Don’t change servers, run several diagnostic tools and measure battery use all at once; the results won’t show which change made a difference.
If your server provides Trojan and valid certificate details, use the complete settings. If certificate errors persist, fix the name or server certificate rather than switching to an unrelated protocol to work around the cause. Once the connection settings check out, investigate routing and the destination service if only certain domains fail. For common app questions, see the FAQ. The checks in this chapter also apply to other protocol combinations that rely on TLS.
HTTP and SOCKS5: authentication, DNS and security limits
General-purpose proxies offer broad compatibility, but the proxy type alone does not mean the connection is encrypted.
What the two general-purpose proxies are designed for
HTTP proxies were designed to forward HTTP requests and can use CONNECT to create a tunnel to a destination. SOCKS5 offers more general connection negotiation and does not require every application to use HTTP semantics. Both are common in existing organizational networks and controlled environments. Their interfaces are clear and widely supported, but server policies determine which destinations and ports are allowed.
An HTTP proxy and an HTTPS destination are two different things. Even when a website uses HTTPS, that does not automatically provide the same protection for the client’s proxy request and authentication details sent to the HTTP proxy. If the proxy entry point itself uses TLS, select the appropriate security type and check the certificate. SOCKS5 negotiation also does not provide general payload encryption. “Supports authentication” does not mean “encrypts all traffic.”
Proxy authentication is not website login
The proxy Username and Password authenticate you to the proxy entry point; they are unrelated to your destination website account. If authentication fails, check the proxy credentials and the authentication methods allowed by the server, not your website login. Some HTTP proxies return a clear authentication error or access denial; other failures may simply close the session. Record the status and affected destination to help distinguish incorrect credentials from an access policy restriction.
An entry point may allow CONNECT only to certain ports or accept requests only from a specified network. If ordinary web pages work but another service doesn’t, that does not necessarily mean Shadowrocket has lost a setting. First confirm the destinations and use cases allowed by the server, then check that you chose the right type. Don’t use an HTTP proxy port as a SOCKS5 port, or assume the same address means both types share authentication. The server must support each negotiation method separately.
Where does DNS resolution happen?
SOCKS5 can send a domain name in the request or send an already-resolved address. Which side performs the lookup also depends on client DNS handling, rule matching and request type. So you can’t assume that all domains are always resolved remotely just because the protocol is SOCKS5. HTTP CONNECT also uses a destination name or address. When troubleshooting, check whether the proxy connection works, whether the destination name resolves correctly and whether the final address is reachable.
If a domain fails but a test using an address can establish the expected connection, check the resolution path first. However, accessing an HTTPS website by address may trigger certificate or virtual-host issues, so a browser error alone doesn’t prove the network is down. Use a test for the same destination that preserves the correct name, and check the diagnostics available in the client. Identify which layer the error comes from before changing DNS or proxy settings.
| Capability | HTTP | SOCKS5 |
|---|---|---|
| Connecting to a destination | Request forwarding or CONNECT, subject to server limits | Connection established through negotiation, subject to server limits |
| Connection encryption | Depends on whether the proxy entry point also uses TLS | The base protocol does not encrypt payloads |
| UDP | Standard CONNECT is not general-purpose UDP forwarding | Requires UDP ASSOCIATE and full forwarding support |
UDP, local networks and iOS requirements
SOCKS5 defines UDP ASSOCIATE, but using it also requires the server to allow datagram forwarding and the client to reach the relay address it returns. Some deployments support TCP only, so authentication can succeed while UDP remains unavailable. A standard HTTP proxy entry point does not automatically support every datagram-based app either. If an audio, video or interactive app fails, check server capabilities before client settings; repeatedly reconnecting won’t help if the required forwarding is unavailable.
On iOS, battery use with these proxies depends mainly on connection count, data volume and network conditions. Simple protocol processing does not mean a shorter route to the destination or no TLS work at the website. For an entry point on a trusted local network or managed by your organization, confirm the access scope and security requirements. On an untrusted network, pay particular attention to whether the connection from the client to the proxy is adequately protected.
Choose based on the requirements of your existing setup: use HTTP settings for an HTTP proxy, or check authentication and datagram support for SOCKS5. Don’t switch the type while keeping the same address and password just to try a different name. If you only need access to specific resources, add the relevant rules in Config and make the handling of other traffic explicit. Rules control what gets routed; they don’t change the proxy’s encryption limits. Check both separately.
WireGuard: keys, tunnel addresses and routing scope
Understand the relationship between the interface and Peer, then check the handshake and tunnel traffic.
A different layer from app-based proxies
WireGuard is designed as a Layer 3 tunnel that carries IP packets inside an encrypted connection. It is not a general-purpose proxy that adds a username and password to every app request, so its configuration differs from Shadowsocks and HTTP. To understand it, consider the local tunnel interface, the remote Peer and the reachable network ranges together. In Shadowrocket, distinguish routing decisions at the app level from the tunnel’s own routes and forwarding requirements.
WireGuard suits existing managed network access or a clearly defined tunneling need, but an established tunnel does not mean the remote side allows access to everything. The server must also handle forwarding, return routes and any required exit configuration. If you see handshake activity but no web response, the outer connection may have reached one stage while the inner address, route or DNS still needs checking. Don’t immediately recreate all the keys.
Private Key, Public Key and Endpoint
The Private Key is the local secret identity; the Public Key lets the peer identify the corresponding device. The remote public key entered on the client must match the actual server, and the server must have the correct client public key. These values are not interchangeable text. If the configuration includes a pre-shared key, it must match on both sides. Remove secrets before sharing a configuration for troubleshooting; never post a complete export or QR code containing credentials publicly.
Endpoint is the reachable remote address and UDP port; it is not the tunnel’s internal address. The tunnel address is used for internal communication after the encrypted connection is established, while Endpoint helps the current network reach the server. If Endpoint is a domain, it must resolve before the outer connection can be established. Network changes can alter the outer route and address mapping, so check whether the handshake recovers—not just whether the app’s switch stays on.
AllowedIPs and address assignment limits
In typical WireGuard configurations, AllowedIPs determines both which destinations are sent to a peer and which source addresses are accepted from it. It is not a general list of app domain rules and cannot replace Shadowrocket rules such as DOMAIN-SUFFIX. A broader address range changes which traffic is expected to use the tunnel; a range that is too narrow may prevent a destination from taking the right path. Preserve the network details provided by the server.
The local tunnel address must also match the details assigned to this client by the server. Combining another device’s address with your own keys can cause address conflicts, incorrect return routing or devices interfering with each other. Consider IPv4 and IPv6 separately: one working does not prove the dual-stack path is complete. If only some destinations fail, check their resolved addresses and whether the relevant address family has a complete path before disabling everything.
How MTU and keepalive affect mobile networks
MTU describes the packet size a path can carry. Extra encryption encapsulation reduces the space available for inner packets; if the path handles fragmentation or size feedback poorly, small requests may work while image loading or uploads stall. Keep the original value before making changes, confirm that the issue is related to network changes, then compare one limited adjustment at a time. There is no single value that works for every Wi-Fi, cellular and nested network path.
Persistent keepalive can maintain a reachable mapping in some address-translation environments; it is not a general speed boost. Frequent intervals can increase background traffic and keep the radio active more often, and may not improve stability when not needed. If the server needs to initiate connections to the client, or idle connections often fail to recover, check this setting against the existing network design. Don’t increase keepalive frequency across multiple entries and then draw conclusions about the protocol’s overall battery use.
WireGuard’s real-world efficiency on iOS also depends on the client implementation, packet-processing path and system scheduling. Don’t assume performance on other operating systems applies directly. Check the handshake, reachability of destinations inside the tunnel, domain access and sustained transfers in turn. If the current network blocks outer UDP, changing internal DNS won’t fix the handshake. If the handshake succeeds but destinations fail, check the inner address, AllowedIPs, server forwarding and return path first.
When your existing setup provides a WireGuard configuration, importing it in full is usually better than copying only Endpoint. When moving to another device, check whether the identity can be reused; the same key on multiple devices can cause conflicts. If you have no working tunnel configuration for ordinary browsing, switching the type in Add Server won’t create one. Use a protocol actually provided by your server.
Hysteria2: QUIC, bandwidth settings and weak-network trade-offs
First confirm that UDP and the server are compatible; then assess whether congestion control suits the network.
The QUIC-based design
Hysteria2 uses a QUIC-based transport that typically relies on UDP and combines authentication and security mechanisms to carry proxy traffic. It targets performance on high-latency, lossy or variable-bandwidth networks, but does not guarantee speed under every network condition. If the current network cannot reliably carry UDP, the transport lacks a necessary foundation. Increasing bandwidth settings or repeatedly changing the password may not address the real cause.
QUIC can manage multiple streams within one connection. Losing data on one stream does not necessarily block delivery on others as it can with a single ordered byte stream. But packet loss still has a cost: retransmissions, acknowledgements and congestion control use network capacity, and streams share the same connection conditions. “Based on UDP” does not mean “unreliable delivery”; the transport layer still handles reliable delivery where needed.
Keep authentication, TLS and protocol version aligned
Address, port and authentication details are basic requirements; TLS names and certificates must also be correct. The protocol version used by the server must match the client type. Similar names do not guarantee compatible message formats. If an imported configuration isn’t recognized correctly, first confirm that its source provides complete details for the current type instead of keeping only the hostname and password. UDP reaching the entry point doesn’t guarantee that security negotiation or app authentication will succeed.
Some deployments specify additional obfuscation settings. These work only when configured to match on both sides; they are not options to enable arbitrarily for extra protection. Authentication secrets, obfuscation secrets and TLS names serve different purposes, so check each one separately. If the server uses a standard certificate, keep verification enabled and check the device time and certificate chain. Don’t disable verification to hide a certificate error caused by an incorrect name or entry point.
Understanding upload and download bandwidth settings
Bandwidth settings contribute to rate decisions in specific congestion-control strategies; they do not increase the capacity of an account. Interpret upload and download from the client’s perspective and follow the server’s instructions. Values set too high may increase queuing, packet loss and retransmissions, affecting responsiveness; values set too low may limit throughput. Some configurations use different congestion-control methods, so whether and how a field takes effect depends on the actual server negotiation.
Estimate available bandwidth from what the current network can sustain over time, not the carrier’s advertised rate or a single peak. Cellular performance changes with signal, load and location, so a fixed value can’t cover every condition. Start with the subscription’s settings and record stable performance. If you adjust a value, change only one direction at a time and compare completion time, responsiveness and failures on the same task—not just a higher momentary reading.
Assess weak-network gains alongside battery costs
On a path with some packet loss, a suitable transport and congestion strategy may improve usable throughput, but can require more packet processing and network activity. Mobile power use depends not only on encryption but also on how long the wireless radio stays active, signal strength and repeated retransmissions. Compare completing the same-sized download quickly with keeping a connection active in the background for a long time; one result cannot represent every use case.
If a network rate-limits UDP or reclaims its sessions, a connection may work at first but fail after sitting idle, or become unstable under heavy load. Compare another access network and check whether the failure happens at the same point. If it occurs on only one network, investigate the outer path first. If authentication fails everywhere, recheck the settings. No protocol should rely on endlessly increasing traffic to compensate for unsuitable network conditions.
Troubleshoot in this order: outer UDP reachability, correct TLS name, successful authentication, correct request routing and destination availability. If a TCP-based protocol works while this entry fails, that points to the outer path as something to compare; it does not prove the account is wrong. On metered networks, remember that repeated tests and retransmissions use data. Avoid long, high-load tests just to chase a peak speed.
If your server offers Hysteria2 alongside other types, evaluate it as an option for specific network conditions and keep a verified alternative for comparison. Choose based on how reliably it completes real tasks: whether pages open promptly, transfers remain stable, the connection recovers after the screen locks and battery use on cellular is acceptable. Note the network used for each test instead of ranking protocols permanently.
Choose by use case, verify routing and troubleshoot by layer
Start with a complete, restorable configuration and make every change a meaningful comparison.
Your existing setup defines the options
If you have only one set of working server details, import it in full using the specified type; there’s no need to rank protocols first. If you have several options, compare initial page loads and repeated visits for everyday browsing; completion time for the same sustained transfer; jitter, UDP support and recovery for real-time apps; and address planning and access scope for managed networks. Compare performance only after confirming each protocol meets the server’s requirements.
For an initial setup, select an imported entry from Home and temporarily use a suitable Global Routing mode to confirm requests are actually sent through the selected proxy. When testing is complete, return to your usual Config mode and verify the active configuration. Don’t add complex rules, multiple policies and automatic connection conditions before confirming the protocol works. The more layers involved, the harder it is to identify what caused a failure.
| What you have | Check first | Test |
|---|---|---|
| Complete proxy subscription | Keep the original protocol, authentication and transport settings | Same web pages and repeated visits |
| Organizational general-purpose proxy | HTTP / SOCKS5 type, authentication and access scope | Specified internal resources and required ports |
| Managed tunnel configuration | WireGuard keys, addresses and routes | Handshake, internal addresses and domain access |
| Variable network with UDP available | Hysteria2 settings and sustained real-world bandwidth | Time on the same task, recovery and battery use |
Rules determine routing, not protocol
The following Shadowrocket configuration example is for learning and contains no server details. Rules are evaluated in order, with more specific conditions first; FINAL handles requests that haven’t matched an earlier rule. PROXY, DIRECT and REJECT specify different actions. Before using the example, check the proxy policies and selected entries in your active configuration. Don’t treat the example domains as a list of required services or replace your existing configuration with the whole snippet.
[Rule]
DOMAIN-SUFFIX,example.com,PROXY
DOMAIN,ads.example.net,REJECT
IP-CIDR,192.168.0.0/16,DIRECT,no-resolve
GEOIP,CN,DIRECT
FINAL,PROXY
DOMAIN-SUFFIX matches a domain and its subdomains; DOMAIN matches a specific domain; IP-CIDR matches an address range. In this example, no-resolve avoids initiating a DNS lookup solely for that address rule; it does not disable DNS across the whole configuration. GEOIP relies on IP geolocation data and cannot precisely represent every website’s business scope. If a request uses the wrong policy, check the match and rule order instead of switching server protocols.
Narrow down the issue by the stage where it fails
If nothing connects, check the local network, server address and port first, then verify the protocol and authentication. For a TLS error, check the device time, name and certificate; for an authentication error, recheck the UUID or password. If the connection succeeds but no destinations work, check server forwarding and DNS. If only some destinations fail, check the rules, destination address family, UDP requirements and destination service. Move on to the next layer only after verifying the previous one, rather than repeatedly overwriting settings you’ve already confirmed.
If small requests work but larger transfers stall, check MTU, fragmentation and path limits. If connections fail after sitting idle, compare session timeout and recovery behavior. If failure follows a switch between Wi-Fi and cellular, check the outer path, DNS cache and connection re-establishment. Don’t explain every symptom as “server failure.” Connectivity Test is a useful signal, but may work differently from real apps; verify with an actual task as well.
Keep repeatable findings and a way to restore settings
Comparison notes should include the network, task, protocol and transport combination, Global Routing mode and results. Change one field at a time and restore its original value if the test fails. Don’t share passwords, full subscription URLs, private keys or QR codes containing credentials in troubleshooting notes. Share only redacted fields and the specific error. Treat exported configuration files as sensitive, and check whether subscription updates will overwrite local changes.
On Demand triggers connections based on conditions; it won’t fix incorrect protocol settings. Verify the connection manually first, then follow the On Demand setup guide to add network conditions. This keeps trigger issues separate from transport issues. Keeping one verified baseline configuration is more useful than enabling many untested adjustments at once. For import steps, return to the tutorial; for common installation and account questions, see the FAQ. This guide explains why each step needs checking.