Cursor and Cline with QuickSilver Pro

Cline and Cursor both let you plug in your own OpenAI-compatible endpoint. QuickSilver Pro is one, so both can use it: one key, every chat model in the catalog, prepaid billing. The two tools differ a lot in how much of the product that key reaches, though. In Cline it is all of it. In Cursor it is the chat and agent panel, with conditions, and not Tab.

You need a QuickSilver Pro API key from the dashboard. New accounts get $0.05 of free credit, which covers the starter models (GPT-6 Luna, Claude Haiku 5.5, DeepSeek V4.1 Flash, Qwen3.8 Flash Next and MiMo-V2.6-Flash); other models need a top-up, from $5.

Both tools need the same two values:

Base URL   https://api.quicksilverpro.io/v1
API key    your QuickSilver Pro key

Cline

VS Code extension

Open the Cline panel, click the settings cog and fill in the provider section:

API Provider   OpenAI Compatible
Base URL       https://api.quicksilverpro.io/v1
API Key        your QuickSilver Pro key
Model ID       claude-sonnet-5-5

The Model ID is any ID from the model list, exactly as written there, with no provider prefix. Under the model configuration options you can also set the context window, maximum output and input/output prices, so Cline's own token and cost display matches your model; the QuickSilver Pro values are on each model's page. The same settings work in Roo Code and Kilo Code, which share Cline's OpenAI Compatible provider screen.

Cline CLI

Cline also ships as a command-line agent (npm install -g cline). Configure the provider once:

cline auth openai-compatible \
  -b https://api.quicksilverpro.io/v1 \
  -k "$QSP_API_KEY" \
  -m claude-sonnet-5-5

This saves the base URL, key and model to Cline's settings, so later runs need nothing else:

cline "test_stats.py fails. Find the bug in stats.py and fix it."
cline -P openai-compatible -m deepseek-v4.1-flash "explain what main.py does"

Note that cline with a prompt runs in act mode with tool auto-approval on: it edits files and runs commands without asking. Use it in a project you can reset, or pass --auto-approve false.

Which models to use in Cline

Prices are per million tokens. Sonnet 5.5 is 75% below Anthropic's list price of $2.00 / $10.00. Cline works with any chat model in the catalog; these two are the ones we ran.

What a small fix cost in Cline

On 2026-10-27 we ran Cline CLI 3.0.74 on a fresh copy of a two-file Python project: average() divided by len(numbers) - 1, and a test caught it. The prompt was "test_stats.py fails. Find the bug in stats.py and fix it with a minimal edit. Do not change the tests. Run the test afterwards to confirm." A run passed if the test passed afterwards. Token counts are what the API reported; the cost is what QuickSilver Pro charged.

Both made the same one-line fix and ran the test. One run per model on an easy bug, so this is an order of magnitude, not a ranking. Cline's system prompt and tool definitions come to about 6,000 tokens per request, before any of your code; on later requests most of it is served from the prompt cache. Sonnet got there in fewer steps; DeepSeek V4.1 Flash took more of them and still cost a fifth as much.

Cline sends a standard Chat Completions request (streaming, with tools), and the CLI's own cost readout matched what QuickSilver Pro billed. Every request also shows up in your dashboard.

Cursor

Cursor's "bring your own key" support is narrower than Cline's, and some of it works differently from what you might expect. What follows is from Cursor's own documentation and from answers by Cursor staff on its forum; we could not run Cursor itself, so we checked the QuickSilver Pro side of each request shape directly against the API instead.

Setup

In Cursor Settings → Models:

1. Turn on OpenAI API Key and paste your QuickSilver Pro key.
2. Turn on Override OpenAI Base URL and enter https://api.quicksilverpro.io/v1.
3. Add the QuickSilver Pro model ID as a custom model, for example deepseek-v4.1-flash, and select it in the chat or agent panel.

What works with a custom key, and what does not

• Chat and agent: yes. Cursor sends Chat Completions requests to your base URL. We checked that QuickSilver Pro answers streaming requests with tool calls for every model in the table below.
• Tab completion: no. Cursor's docs: "Custom API keys only work with chat models. Tab completion continues using Cursor's built-in models."
• Requests go through Cursor's servers. Also from the docs: the key "is sent to our backend with every request because all requests are routed through Cursor's servers for final prompt building." QuickSilver Pro is a public endpoint, so this works; it does mean Cursor sees your key on every request.
• Model IDs starting with claude- do not reach QuickSilver Pro this way. Cursor staff explain that Cursor picks a key by the start of the model name, and a name beginning with claude- goes to the Anthropic key slot, not the OpenAI override. So QuickSilver Pro's Claude models are not usable through this setting.
• GPT model IDs may break. For GPT-family models, Cursor sends a Responses API body to the /chat/completions URL (acknowledged by Cursor staff, no fix date). QuickSilver Pro, like other Chat Completions endpoints, rejects that body with messages is required: a non-empty array of {role, content} objects. We could not test which GPT IDs Cursor treats this way, so we suggest the open models below.
• The override is global. It applies to every OpenAI-family request, including Cursor's own models, and Cursor's Composer models fail with "This model does not support custom API keys" while it is on (forum). Turn the OpenAI API Key toggle off to go back to Cursor's models; your base URL stays saved.
• A paid Cursor plan is needed for custom keys, according to Cursor staff on the forum.

Which models to use in Cursor

Prices are per million tokens. None of these IDs start with claude-, gemini- or gpt-, so they stay clear of the routing issues above. If you want Claude or GPT models with your QuickSilver Pro key, Cline above, Claude Code and Codex CLI all reach them without these limits.

Troubleshooting

• 401 or "invalid key". Check that the key field holds your QuickSilver Pro key and the base URL is exactly https://api.quicksilverpro.io/v1, with no /chat/completions on the end; both tools add that themselves.
• "Unknown model or endpoint". The model ID is case-sensitive and has no provider prefix: deepseek-v4.1-flash, not deepseek/deepseek-v4.1-flash.
• Cursor: messages is required. You hit the GPT-model issue above. Pick a model from the Cursor table instead.
• Cursor: your key seems ignored for a Claude model. That is the claude- routing above; the request never reaches QuickSilver Pro.

Using OpenCode or Crush instead? See Use QuickSilver Pro in OpenCode and Crush. Other tools are on the integrations page.