Cline with the Claude API — base URL setup, Plan/Act model routing, real costs

Cline is an open-source autonomous coding agent for VS Code, and it does not ship a model — you bring your own API key. This guide is the exact settings for running Claude in Cline through one sk-kn- key at about 60% under Anthropic's list price on the mainline models: both provider routes, how to exploit Plan/Act mode to cut the bill, what a real session costs, and an honest account of what you give up versus calling Anthropic directly.

Why Cline needs an API key

Every action Cline takes — reading a file, proposing a diff, running a terminal command, reading the output and trying again — is a model call billed per token. A claude.ai Pro or Max subscription does not cover this: chat plans and API access are separate products, and Cline talks to the API. So the question is not whether to pay per token but at what rate, and because an agentic loop resends the growing task context on every step, the per-token rate is the single biggest lever on what a session costs. That is the whole case for a gateway here: pay-as-you-go from a $10 top-up, one key that also reaches Gemini and GPT, and a balance that never expires.

Setup — the three fields

Open the Cline panel, click the settings gear, and set the API provider to OpenAI Compatible. Three values:

API Provider   OpenAI Compatible
Base URL       https://api.kunavo.com/v1
API Key        sk-kn-...              # create at kunavo.com/app/keys
Model ID       claude-sonnet-4-6      # or claude-sonnet-5 / claude-opus-5

Create the key at the dashboard after signing up and topping up — it is shown once, so store it immediately. Key management, rotation and spend limits are covered in the Claude API key doc. If Cline asks for model metadata such as the context window or output cap, take the numbers from the model page rather than guessing — an understated context window makes Cline truncate earlier than it needs to.

Before letting it edit anything, ask it something small (“explain this file”) to confirm the wiring. A 401 at this point is almost always the key; a 404 is almost always the base URL or a model slug that does not exist — Kunavo matches slugs exactly and does not alias date-suffixed names, so claude-sonnet-4-5-20250929 is a 404 while claude-sonnet-4-6 resolves.

Plan mode and Act mode — the cheapest split

Cline's defining feature is that it separates Plan (read the codebase, ask questions, propose an approach) from Act (make the edits, run the commands), and it lets you configure a different model for each. That maps almost perfectly onto cost, because the two modes have very different token profiles:

• Plan is short and reasoning-heavy. A handful of turns where model quality decides whether the next twenty steps are productive or wasted. Put claude-opus-5 or claude-sonnet-5 here — it is a small share of the tokens, so the stronger model costs little in absolute terms.
• Act is long and mechanical. Many steps, each resending context. This is where the per-token rate compounds, so claude-sonnet-4-6 is the value default.

The worked numbers are in the cost section below: planning three steps on Opus 5 and executing seventeen on Sonnet 4.6 costs about $0.86 against $0.74 for the all-Sonnet run — roughly 16% more for materially better planning, and far less than running the whole task on Opus.

The native Anthropic route (and why you might want it)

Cline also ships a native Anthropic provider with a custom base-URL field. Pointing that at Kunavo uses the native Messages API (/v1/messages) instead of the OpenAI-compatible endpoint:

API Provider   Anthropic
Base URL       https://api.kunavo.com   # origin only — the client appends /v1/messages
API Key        sk-kn-...
Model ID       claude-sonnet-4-6

Note the base URL differs between the two routes. The OpenAI-compatible provider wants https://api.kunavo.com/v1; the Anthropic provider wants the bare origin https://api.kunavo.com, because that client appends /v1/messages itself. Pasting the /v1 form into the Anthropic provider produces a 404 that looks like a missing model.

What the native route buys you is prompt caching: cache_control passes through untranslated, and cached input bills at 10% of the input rate (see the prompt-caching doc). On a loop that resends a stable prefix every step, that is the largest saving available — larger than any model swap. Either route works; the OpenAI-compatible one is simpler to start with, and switching later is a settings change.

Which Claude model to set

Switching is a one-word change to the model field — same key, same provider config, same endpoint. Full per-model rates against Anthropic's official list are in the Anthropic Claude API price list, and the live catalog is on the pricing page.

Cline on Kunavo vs calling Anthropic directly

The honest comparison, including where going direct is the better call:

Read that capacity row as written. If your team needs a guaranteed quota, a contractual SLA, or a procurement relationship with the model vendor, go direct — that is what you are paying the difference for. For an individual developer or a small team running Cline, the trade is usually the other way around.

What a Cline session actually costs

Agentic tools are token-hungry by design: each step resends the system prompt, the task history and fresh file context. Realistic numbers at Kunavo rates:

The math, runnable:

# Kunavo Claude rates (USD per 1M tokens): (input, output)
RATES = {
    "claude-haiku-4-5":  (0.40, 2.00),
    "claude-sonnet-4-6": (1.20, 6.00),
    "claude-opus-5":     (2.00, 10.00),
}

def step_cost(model, in_tokens, out_tokens):
    i, o = RATES[model]
    return in_tokens / 1_000_000 * i + out_tokens / 1_000_000 * o

# One Act-mode step: Cline resends the task context plus the files it read.
print(step_cost("claude-sonnet-4-6", 25_000, 1_200))       # -> $0.0372
# A realistic 20-step task (edit, run, read the error, fix, repeat):
print(20 * step_cost("claude-sonnet-4-6", 25_000, 1_200))  # -> ~$0.74
# The same task planned on Opus 5, executed on Sonnet 4.6:
print(3 * step_cost("claude-opus-5", 25_000, 1_200)
      + 17 * step_cost("claude-sonnet-4-6", 25_000, 1_200)) # -> ~$0.86

Keeping the bill down

1. Start new tasks instead of extending one forever. Cline resends the whole conversation each step, so a long-running task grows quadratically in cost. A fresh task is a fresh, small context — this is the highest-leverage habit on the list.
2. Split Plan and Act across models. Covered above: it is a settings field, not a discipline problem.
3. Use the native Anthropic route for caching. Cached input at 10% of the input rate is the biggest structural saving on a loop that resends a stable prefix.
4. Give the editor its own spend-limited key. Create a separate key in the dashboard, set a monthly spend limit on it, and watch usage — a week of real agentic coding is the only honest estimate of what this costs you. The limit is checked before each request and resets with the calendar month, and the key is revocable on its own, so stopping the editor is one action rather than rotating everything.
5. Still choosing between the two tools? Cline vs Claude Code compares the billing shapes rather than the feature lists — and both run against this same endpoint, so the choice stays reversible.

FAQ