iTranslated by AI

The content below is an AI-generated translation. This is an experimental feature, and may contain errors. View original article
🤖

Running a Coding Agent in the Browser with code-server and Cline

に公開
2026/01/14

code-server is an OSS that allows you to run VS Code in a browser.
By building code-server on Docker Desktop and installing Cline, we can run a coding agent directly in the browser.
https://github.com/coder/code-server


Environment

  • Windows 11 Pro 25H2
  • Docker Desktop v4.52.0
  • code-server
  • Cline
  • Tailscale

Search and Run code-server from Docker Hub


Search for codercom/code-server.


You will find the official image, so select it.


Select the version and click the Run button.
The image will be pulled at this point.


Since 8080 is already in use in my environment, I will specify 8100 this time.

Once you Run it, the container will start.

Access from the Browser


Access localhost:8100 in your browser. If it has started correctly, you will see a password input screen.
The password is listed in the following file:
/home/coder/.config/code-server/config.yaml

Additionally, you can make it accessible without a password by changing auth: password to auth: none and restarting, as shown below:

config.yaml
bind-addr: 127.0.0.1:8080
- auth: password
+ auth: none
- password: xxxxxxxxxxxxxxxxxxxxxxxx
cert: false

Installing the Japanese Language Extension


First, let's install the Japanese language extension.

In a standard VS Code, installing it is sufficient to switch the language after a restart, but in code-server, explicit settings are required.

Press the F1 key, type Configure Display Language, and explicitly select Japanese.

Installing Cline


Install it from the Extensions view.

Enabling HTTPS with Tailscale

Simply installing Cline is not enough for it to function correctly.

https://github.com/cline/cline/issues/1001
As noted in the issue above, a phenomenon occurs where "a blank page is displayed and Cline does not work."

As mentioned in the issue comments, if you check for errors in the browser's developer tools, you will find that the cause is accessing via http.

Since it will function correctly once HTTPS is supported, we will use Tailscale to resolve this.

If you haven't set up the Tailscale environment yet, please refer to the following article:
https://zenn.dev/sion_pn/articles/f3bc98a32ed3ab

After setting up the environment, run:

tailscale serve --https=8100 8100

While the above command is running, you will be able to access it via HTTPS at https://<device-name>.<tailnet-name>.ts.net:8100.

Cline Initial Setup


Since I want to run it with a Local LLM, I select Bring my own API key.


I am running a Local LLM using LM Studio, so I specify LM Studio as the API Provider and explicitly set the URL.
Once the connection is established, you will be able to select from a dropdown list of available models; for this instance, I selected google/gemma-3-27b.


That is all.
Local LLMs are great because you can try out coding agents without worrying about usage fees.

Discussion