Cline in JetBrains: Installing in IntelliJ IDEA and PyCharm

Cline started as a VS Code extension, but it can now be installed in JetBrains products too. If you write code in IntelliJ IDEA, PyCharm or PhpStorm and want an agent that asks for approval of edits right in the IDE window, Cline in JetBrains fits. Below: installing the plugin, connecting to the Zukko gateway, tuning auto-approve and a fallback through the ACP protocol if the plugin does not suit you.

Cline JetBrains: installing the plugin

According to Cline's documentation, the path for JetBrains is:

1. Open Settings → Plugins → Marketplace.
2. Find Cline and click Install.
3. Restart the IDE.
4. Open the panel via View → Tool Windows → Cline.

The plugin is also available on the JetBrains Marketplace↗ page. Menu item names depend on the IDE version and the new UI shell, so if you cannot see a menu entry, use the settings search.

The list of editors where the Cline extension installs includes VS Code, Cursor, Windsurf, VSCodium, Antigravity and JetBrains in the documentation. So the JetBrains version is an official member of the family, not a third-party build.

Connecting to the Zukko gateway

Cline works with any OpenAI-compatible provider, and the Zukko gateway fits. You need three values.

1. Base URL: https://api.zukko.pro/v1. Clients on the OpenAI format use the address with /v1.
2. Key: the fetch_... string created in the dashboard↗. The full key is shown once.
3. Model id from GET /v1/models, for example claude-sonnet-5-5 or gpt-6.1-sol.

In the Cline panel click the settings icon (the gear), choose OpenAI Compatible in the API Provider field and paste the Base URL, key and model id. Below there is a block of model parameters: max output tokens, context window size, image support and prices. Cline needs these values to count and trim context correctly, so fill them in from the table: for example, claude-sonnet-5-5 accepts images and has a 1M context, while deepseek-v4.1-flash does not accept image input. If the interface offers a connection check button, press it. The general scheme is described in the Zukko documentation↗ (the VS Code extensions section, which includes the Cline setup) and in the article on connecting Cline↗.

We checked the field names against Cline's documentation for OpenAI-compatible providers. In a specific version of the JetBrains plugin the layout may differ slightly, and if you see different labels, go by meaning: address, key, model.

How the JetBrains version differs from VS Code

Let's be honest: we did not find a detailed table of differences in Cline's documentation. Here is what is known:

• the configuration is shared: provider settings, rules and skills live in ~/.cline/, and project rules live in .clinerules/ or .cline/rules/, read by both IDE versions and the console one;
• the interface lives in a JetBrains tool window instead of the VS Code side panel;
• the feature set between editors can diverge because of different releases: some features appear in one shell earlier. See the plugin's release notes for the change list.

Verify the feature you need (for example MCP server support or the built-in browser) in the plugin itself rather than from the VS Code description.

Auto-approve: set it before the first task

Cline asks permission for every action until you enable auto-approve. The documentation lists the options by category: reading project files, reading outside the project, editing project files, editing outside the project, safe commands, all commands, browser, MCP servers. An important detail: "all files" and "all commands" only work on top of the base toggle. If the base toggle is off, the extended one does nothing.

A sensible starting setup for the IDE:

• enable reading project files;
• keep edits of project files on confirmation until you get used to the agent's style: changes open in the diff window and are convenient to accept in a batch;
• "execute safe commands" can be enabled, "all commands" better left off;
• leave external files and the browser alone.

For long tasks the Plan and Act modes help: the agent first drafts a plan, you read it, then switch to execution. That is cheaper than fixing a failed attempt.

What it costs and the cache catch

Prices as of October 18, 2026. Say a task makes 20 requests and each contains on average 100,000 input tokens (the context grows, Cline sends the whole accumulated conversation), so 2M input tokens in total, with 40,000 output tokens.

claude-sonnet-5-5 through the OpenAI format ($0.4 input, $2 output): 2 × 0.4 + 0.04 × 2 = 0.8 + 0.08 = $0.88.

The catch is that in the OpenAI format the cache discount does not apply to Claude: all input is billed at the input price, however well the repeated context is cached. For GPT models, cache reads are visible in both formats. Suppose that for gpt-6.1-sol (input $0.27, cache read $0.013, output $1.32) 90% of the input hit the cache; that is an assumption, not a guarantee. Then: 0.2 × 0.27 + 1.8 × 0.013 + 0.04 × 1.32 = 0.054 + 0.0234 + 0.0528 = $0.1302.

Do not conclude from this that one model is better than another: they differ in strength index (48 for claude-sonnet-5-5, 52 for gpt-6.1-sol) and in behavior on agent tasks, and you should test that on your own project. If the cache discount for Claude matters to you, look at the console Claude Code↗, which works in the Anthropic format, or at the Anthropic provider in Cline with its custom base address option: for Anthropic-format providers the gateway expects the address without /v1 (https://api.zukko.pro), and if you get a 404, check that suffix first. Current prices: GET https://api.zukko.pro/v1/models or the Models page in the dashboard.

Fallback: Cline as an ACP agent

If the plugin does not start or you need an unusual setup, Cline has an ACP protocol mode: the console client cline --acp can act as an agent inside JetBrains AI Assistant. According to Cline's documentation, in the AI Chat window you choose Add Custom Agent, which creates ~/.jetbrains/acp.json, and add an entry to it:

{
  "agent_servers": {
    "Cline": {
      "command": "cline",
      "args": ["--acp"],
      "env": {}
    }
  }
}

The console client installs with npm install -g cline, and credentials from cline auth are picked up automatically. If the IDE cannot find cline, give the full path from which cline. In ACP mode nothing is approved automatically by default. For a comparison with the IDE's built-in assistant see the note on JetBrains AI Assistant↗.

Common problems

401 after entering the key. Make sure the key field has no spaces and that the Base URL ends with /v1.

404 model_not_found. A typo in the id: take the exact name from GET /v1/models.

403 model_not_allowed_for_key. The model is not on the key's allowed list.

402. The balance ran out. Top up from $1 via SBP (the payment page adds an 8.5% fee), a MasterCard card or cryptocurrency.

429 key_daily_limit_reached. The key's daily limit is exhausted. For the IDE it is handy to create a separate key with a dollar limit so an accidental loop does not burn the balance.

FAQ

Is there a Cline for JetBrains?

Yes, the plugin installs from the JetBrains Marketplace and opens as a tool window. Cline's documentation lists it among supported editors.

Which Base URL should I use in Cline?

For an OpenAI-compatible provider: https://api.zukko.pro/v1.

Can I use models other than Claude?

Yes. Any model can be called in any format: GPT, Gemini, DeepSeek, Qwen and GLM are available on the same key. Take the id from GET /v1/models.

What should I choose: Cline or the built-in JetBrains assistant?

It is a matter of habit and tasks. Cline gives an agent with a plan and edit approval on your own provider, while the built-in assistant is integrated deeper into the IDE. Try both on a real task.