All posts
·4 min readclinevscodetutorial

Cline + Custom Base URL: Same VS Code Agent, 80% Lower Bill

Cline (formerly Claude Dev) made plan-and-act a one-click button. It also makes your Anthropic invoice climb fast. Here's how to point Cline at a cheaper Anthropic-compatible endpoint without losing a single feature.

Cline is the best autonomous coding agent in VS Code. Period.

If you have used Cline (the fork of Claude Dev), you already know. The plan/act split, the file-tree awareness, the diff approval flow — it is the closest thing to Claude Code without leaving the editor. It is also one of the most expensive, because every plan/act loop reads files, sends them to Claude, and gets fresh context back.

Cline's own documentation warns about token usage. Heavy users routinely report $5-15 per day in Anthropic spend. At that pace you cross $200/month before mid-month — and Cline doesn't have a built-in way to throttle.

There is a way: Cline supports custom Anthropic base URLs. You can point it at any Anthropic-compatible endpoint. Like, say, claudeapi.cheap, which gives you the same Claude models at 70-80% off.

Where to set it

The setting is in Cline's UI. Open VS Code:

1. Click the Cline icon in the left sidebar.

2. Click the gear icon (settings) at the top of the Cline panel.

3. Under API Provider, choose Anthropic.

4. Paste your sk-cc-... key into Anthropic API Key.

5. Tick Use custom base URL.

6. Paste https://claudeapi.cheap/api/proxy into the URL field.

7. Save.

That is the entire setup. Cline is now sending every request through claudeapi.cheap.

What still works

In case you are nervous: everything.

  • Plan mode (Sonnet by default). Same speed, same plan quality.
  • Act mode (Sonnet or Opus). Same diff format, same approval flow.
  • File reading and writing — that's all client-side in Cline. We don't touch files.
  • Prompt caching — forwarded through. Cline's heavy reuse pattern (same files re-sent) gets cache hits at 0.1× input cost, exactly like Anthropic direct.
  • Tool/function calls — fully preserved.
  • Streaming — fully preserved.
  • Error handling — Cline's retry logic and rate-limit detection work unchanged because we forward the same status codes.

    • The only thing that changes is the bill.

    How much you save

    Cline's heaviest token consumer is plan mode running on Claude Sonnet 4.6. Official Sonnet pricing is $3 input / $15 output per 1M. Through claudeapi.cheap Pro (80% off), it is $0.60 input / $3.00 output. A typical 50k-token plan that costs $0.30 directly costs $0.06 through us — five-sixths of the bill, gone.

    For a developer running Cline for ~2 hours/day on a real codebase:

    | Plan | Anthropic direct | claudeapi.cheap (Pro) | Saved |

    |---|---|---|---|

    | 1 hour / day | $90 | $18 | $72 |

    | 2 hours / day | $200 | $40 | $160 |

    | All-day (5+ hr) | $500 | $100 | $400 |

    The Pro plan is $19, paid once, lifetime. It pays itself back in week one for any real Cline user.

    When to use Sonnet, Opus, or Haiku

    Since every model is now 80% cheaper, you can be more generous with model choice. Quick rules of thumb:

  • Sonnet 4.6 — Cline's recommended default. Best speed/quality balance for plan and act.
  • Opus 4.7 — Save for hard refactors, architecture decisions, or when Sonnet's output drifted. At $0.60/$3 (Pro), Opus is now cheaper than Sonnet was at official rates.
  • Haiku 4.5 — Excellent for cheap planning loops on small files, file scans, or sub-agent fan-out. $0.12/$0.60 per 1M.

    • Cline lets you flip models per-task in the UI, so you can A/B without restarting.

    Verifying it worked

    After saving the settings, run any small Cline task. In your claudeapi.cheap dashboard, look at usage logs — within ~5 seconds of Cline finishing, you should see a row with the model and token count matching what Cline reports. If you see it, you are routing through us. If you don't, the URL or key didn't save (re-open the Cline settings and check).

    Cline writes the final response to your editor exactly as before. The proxy is invisible.

    Other tools, same key

    claudeapi.cheap is one `sk-cc-...` key for three vendors — Claude, GPT-5, and Gemini. If you also run Claude Code, Aider, OpenClaw, or Cursor, the same key works everywhere via per-tool guides on our Tools page.

    Sign up

    Free Basic plan gets 70% off and 200 RPM — enough to test Cline at full pace for a few days. If it sticks, Pro at $19 lifetime unlocks 80% off and 500 RPM permanently.

    With Cline, the bill grows with your productivity. With claudeapi.cheap, the productivity stays — only the invoice shrinks.