First-Time Clash Connection: Choose a Node, Test Latency, and Verify Your Proxy
After importing a subscription, choose a node, test latency, enable the system proxy, then check your IP and connection logs to confirm traffic is routed through the proxy.
Enable the subscription and select the policy group you want to use
Seeing a subscription in the list only means the client has saved that configuration. Open the client’s configuration page, select the configuration you just imported, and make sure it’s enabled. Then open the proxy page and check that policy groups and nodes are available. If a group appears but contains no usable nodes, check the configuration update or core loading logs first. A configuration parsing failure and a remote proxy list that hasn’t loaded require different fixes; repeatedly toggling the system proxy won’t make nodes appear.
- Check the operating mode. This guide uses Rule mode as its baseline. In Rule mode, rules determine whether a request goes through a policy group, connects directly, or is rejected. Global mode uses the global policy selection, while Direct mode sends ordinary proxy requests entering the core directly. Switching to Global mode does not automatically route traffic from every app.
- Find the group used by the matching rule. The group name is defined by the subscription and may not be called “Proxy Selection.” If you can inspect the active configuration, check
rulesand find the first rule that matches your test domain. Domain rules can usually be matched against the hostname; for IP-based rules or rule providers, use the actual match shown in the connection panel. - For a controlled comparison, use a group that lets you choose manually. A
selectgroup lets you pick a node and may also contain another group. If you select an automaticurl-testgroup, that group’s latency tests still determine the node used. To change only one node between two tests, manually select a specific node along the path used by the matching rule, and note its name.
Test node latency and shortlist two candidates
Find the batch latency test in the proxy page. If possible, run two consecutive tests on the same network and with the same test URL, then choose one primary node and one backup. The client’s manual test and the scheduled probes for a url-test group in the configuration may use different URLs or run at different times, so their results don’t have to match.
| Candidate node | Test 1 | Test 2 | What to use for the first connection |
|---|---|---|---|
| A | 92 ms | 101 ms | Both tests succeeded; try this one first |
| B | 64 ms | Timed out | Test again; don’t choose it just because it had the lowest value |
| C | 148 ms | 153 ms | Both tests succeeded; use it for comparison |
The latency in milliseconds is the time recorded when the client sends a request to the probe address through the selected node; the exact timing depends on the implementation. It is not download bandwidth, and it doesn’t guarantee that another domain is reachable. If every node times out, first check that your local network is working, then review the test URL and timeout settings. If only one node times out repeatedly, switch to one that succeeds consistently and continue testing; there’s no need to change DNS and the operating mode at the same time.
Enable the system proxy and check the local listening port
Leave TUN off for now. Turn on the client’s “System Proxy,” then send a request from a browser that follows system proxy settings. This toggle typically updates the system proxy configuration; the local core listening on your device is what actually receives requests. For example, on Windows 11, if the client sets a manual proxy, go to Settings → Network & internet → Proxy → Manual proxy setup to check the address and port. If the client uses an automatic configuration script, check Automatic proxy setup on the same page instead of enabling a manual proxy as well.
- If the configuration uses
mixed-port: 7890and listens on the local loopback address, you can use127.0.0.1:7890as the HTTP proxy address. Use the listening port shown by the current client; desktop clients may override the port in the subscription. portis the HTTP proxy listening port,socks-portis the SOCKS proxy listening port, andmixed-portaccepts both. Don’t enter a SOCKS-only port in the HTTP proxy field, or use theexternal-controllerport for the control panel.- If your system already has a corporate proxy or automatic configuration script, note its settings before making changes. If the client says the proxy is enabled but your browser still isn’t using it, check whether an existing auto-config, the browser’s own proxy settings, or an extension is changing the route.
Verify the proxy using your exit IP and connection logs
Your exit IP tells you which address the lookup site sees; connection logs show how the client handled the request. Check both for the same visit. In Rule mode, a lookup site going through a proxy doesn’t mean other sites use the same node.
- Record a baseline. Without changing your existing network settings, turn off the system proxy and TUN for this test. In the same browser, open a trusted IP lookup site and note the address and whether it reports IPv4 or IPv6. Don’t compare an IPv4 result from one test directly with an IPv6 result from another.
- Make the request again. Enable the system proxy and confirm the node you selected is still available. Open the client’s “Connections” panel, then visit the same lookup site in a new browser window. If the browser is reusing an existing connection, quit and reopen it; refreshing the page may not create a new connection.
- Check the rule and outbound route for the same host. Find the connection using the lookup page’s hostname and the time you visited it. Connection details in mihomo clients typically show the matched rule and route, though labels vary by client. If the route ends at your selected node, the request used that proxy outbound; confirm it against the page response and exit address. If the page gets the IP from a separate API domain, check the request that returned the IP.
| What you see | What it means and what to do next |
|---|---|
| Lookup succeeded; route ends at the selected node | This request went through the proxy. If the address matches your baseline, first check the address family used by the lookup, any cached result, and the host that returned the actual IP. Don’t dismiss the connection log based on the IP alone. |
Connection logged; outbound is DIRECT | The request reached the client, but a rule sent it directly. Check the matched rule, then compare using a test domain you know should go through the proxy. |
| No matching connection log | First check whether the short-lived connection disappeared from the active list. Watch the panel while refreshing, or check the logs available in your client. If it’s still missing, recheck the system proxy address and port, along with the app’s proxy settings. |
| Route points to a node, but the page doesn’t load | The route only shows that an outbound was selected; it doesn’t mean the request completed. Check the logs for a DNS resolution failure, connection timeout, or other error, then try the same domain again with your backup node. |
If the browser can’t connect, test the proxy port separately
On Windows, if you can run curl.exe, use the command below to send a single request explicitly through the local HTTP proxy. This separates browser settings from the local listener. The example uses 7890; replace it with the client’s actual HTTP or mixed port. The test domain is example.com. If it matches a direct-connection rule, this command only tests the local proxy entry point and direct route; it does not confirm that traffic can exit through a node.
curl.exe --noproxy "" --proxy http://127.0.0.1:7890 --head --max-time 15 https://example.com/
If you get a connection error for 127.0.0.1:7890, first check that the core is running and the port is correct. If you see 200 Connection established followed by a timeout, that status only means the proxy accepted the HTTPS tunnel; check the target site’s final response and the client logs. If the command gets a response from the target site but the same domain fails in your browser, check the browser’s proxy settings before changing subscriptions.
For the first check, record four things: the active configuration, the matched policy group, the actual outbound node, and the test domain result. When retesting, keep the domain, browser, and network the same, and change only one thing at a time: the node or the traffic-routing method. When sharing troubleshooting details, include only the relevant rule, error message, and port. Redact subscription URLs, credentials, and any addresses you don’t need to share.