DeepSeek Harness URL won't open: 127.0.0.1:3080 variants

A copied DeepSeek Harness address that won't open is not usually a broken service but a corrupted address: a full-width colon, a missing http://, stray spaces, or the terminal's dsh web: prefix pasted along with it (source). This guide fixes that branch first, then gives the order to check when the address itself is fine.

DSH plugin overview: tell a bad address from a service that never started

There are only two families of causes: the address string is invalid, or the address is valid but the service or port has a problem. They look different and you handle them in a different order:

One line: ten seconds spent verifying the address beats repeatedly restarting the service.

Why a copied DSH plugin address won't open: four common typos

Every character in this address has to be exact — a single misplaced character makes the browser fail to parse it. Compare the four usual typos one by one:

1. A full-width colon. The address is written as 127.0.0.1：3080 (full-width ：) instead of 127.0.0.1:3080 (ASCII :). Expected: it opens once the full-width colon is replaced. With a CJK input method the colon is full-width by default, which is the easiest way to hit this while typing by hand.
2. A missing http:// scheme. Only 127.0.0.1:3080 was written. Expected: most browsers can complete it to http, but that completion is unreliable — the address may be treated as a search term or reached over https and fail; write it out explicitly to be safe.
3. Stray spaces or a line break. When copied from a chat window or a document, a space ended up inside the address, or a line break at the end. Expected: a space in the address bar means the address is invalid — delete the extra whitespace.
4. The terminal prefix copied too. The terminal prompt reads dsh web: http://127.0.0.1:3080, and pasting the whole line drags the dsh web: part in with it. Expected: keep only the part after http://. dsh web: is a prompt prefix, not part of the address.

What the four share: the address string is invalid or has extra content, so the browser never even tries to connect — which is why restarting the service changes nothing.

What the correct DSH plugin address looks like and how to verify it

The canonical form is http://127.0.0.1:3080 — scheme, loopback address, ASCII colon, default port. Verify it in four steps:

1. Check the scheme: the address should start with http://. Expected: the scheme is present and the string does not begin with a digit.
2. Check the colon: the character between 127.0.0.1 and 3080 must be an ASCII :. Expected: the colon is narrow and centred; a full-width ： is wider and sits left of centre, so the two are easy to tell apart at a glance.
3. Check for extra characters: confirm there is no space, no line break, and no prefix such as dsh web:. Expected: the address is continuous in the address bar, with no whitespace.
4. Check the port number: the default is 3080; if you started with a different port (--port), this has to change with it. Expected: the port in the address matches the port in the startup log.

Comparison table (wrong form → correct form):

DSH plugin address is right but it still won't open: troubleshoot in order

If the address is written correctly and it still fails, the problem is on the service-and-port side — work down the list. Four steps:

1. Confirm the process is still running. Go back to the terminal window that started dsh web and see whether the service is alive. Expected: the terminal still shows running logs; if the window was closed or the process exited, run npx @deepseek-ai/dsh web or dsh web again.
2. Confirm the port is not taken. If another program holds the port, the service may fail to start or move to another port. On macOS and Linux use lsof -i :3080; on Windows use netstat -ano | findstr :3080. Expected: if the holder is not the current service, release it, or switch with dsh web --port 8080 and update the address to match.
3. Check whether it is the authentication prompt. If the page connects but shows authentication required, that is an expired browser session, not an address problem. Expected: restart the process and reopen the freshly printed tokenised address once, as explained in dsh web authentication and port.
4. Rule out proxy and browser interference. A system-wide proxy that forwards 127.0.0.1 as well will intercept the local address, and individual browser caches can block it too. Expected: add the local address to the direct-connect rules, or retry in a private window; the full checklist for ports and refused connections is in 127.0.0.1:3080 unreachable.

Once the address works, the next step is connecting it to the plugin ecosystem: use DSH Plugin Hub to browse, one-click install and manage DeepSeek Harness plugins instead of typing commands by hand.

DSH plugin notes

These are the common DSH plugin address gotchas worth remembering. Item by item:

1. Glance at the colon after copying: the full-width ： is the most frequent typo, especially under a CJK input method.
2. An address with a prefix is an address with garbage: dsh web: is only a prompt label — delete it before pasting.
3. Write the scheme in full: do not rely on the browser's autocompletion; an explicit http:// is the safest.
4. Changed the port? Change the address: after switching ports with --port, the port in the address has to follow, or you connect to an empty port.
5. Separate authentication from refused connections: one is an expired session, the other means nobody is listening — the fixes are completely different.

Sources: dsh CLI README, dsh-cmdline README, dshplugin/dsh-plugin-hub