OpenCode 被迫移除 Claude：Anthropic 畫下開源 AI 工具的訂閱紅線 | OpenCode Forced to Drop Claude: Anthropic Draws the Line on OAuth Token Abuse
By Kit 小克 | AI Tool Observer | 2026-03-29
🇹🇼 OpenCode 被迫移除 Claude：Anthropic 畫下開源 AI 工具的訂閱紅線
三月底，一個在 GitHub 上累積超過 12.5 萬顆星的開源專案悄悄推送了一個讓社群嘩然的 commit：OpenCode，這個被開發者廣泛使用的終端 AI 編程助手，宣布移除所有 Claude 模型的整合支援。原因只有一行：「anthropic legal requests」。
OpenCode 是什麼？
OpenCode 是一個以 Go 語言撰寫的開源命令列 AI 編程助手，功能類似 Claude Code 或 GitHub Copilot，但完全開源、可自行架設，並支援多個 AI 模型。Claude 因為智能表現較強，成為許多用戶的首選。
問題在於：許多用戶並非透過付費 API 金鑰存取 Claude，而是利用 Claude Pro/Max 訂閱帳戶的 OAuth Token，讓 OpenCode 偽裝成 Claude Code 官方客戶端對 Anthropic 伺服器發送請求。簡單說，就是用每月 20 美元的訂閱方案，當成無限制的 API 使用。
Anthropic 為何出手？
這並非突然的決定。完整時間軸如下：
- 2026 年 1 月 9 日：Anthropic 啟動伺服器端封鎖，第三方工具使用 OAuth Token 開始收到錯誤訊息：「This credential is only authorized for use with Claude Code」。
- 2026 年 2 月 19 日：Anthropic 正式更新服務條款，明確禁止 Free/Pro/Max 訂閱計畫的 OAuth Token 用於第三方工具或 Agent SDK。
- 2026 年 3 月：Anthropic 發出法律要求，OpenCode 隨即提交 commit 移除所有 Claude 整合。
從商業角度看，這合乎邏輯：用 $20/月 訂閱「薅」API 流量，嚴重侵蝕了 Anthropic 的 API 收入，也對伺服器基礎設施造成壓力。
開發者的憤怒與背後問題
這個移除 commit 在 GitHub 上收到 40 個踩，只有 4 個讚。HackerNews 討論超過 398 則留言，兩派意見分明：
- 支持 Anthropic：訂閱條款本來就不允許這類用法，使用者「鑽漏洞」本就應該被封鎖，這是正常的商業保護。
- 批評 Anthropic：Claude Code 官方本身也是命令列工具，Anthropic 卻只授權自家工具使用 OAuth 認證，明顯有反競爭之嫌，不利於開源生態發展。
更深層的問題是：開源 AI 工具生態，本質上仍然依賴商業模型供應商的善意與容忍度。當同一家公司既是模型提供者、又是競爭性工具供應商時，生態系統的獨立性就極為脆弱。
對開發者的實際建議
- 若你的工具鏈依賴 Claude，務必使用正式 API Key，而非 OAuth Token。
- 評估多模型策略（OpenAI、Google Gemini、本地 Ollama），降低對單一供應商的依賴風險。
- 密切關注 Anthropic 服務條款異動，尤其是 Agent SDK 的使用限制部分。
值得一提的背景：Anthropic 的付費訂閱用戶今年確實翻倍成長，每日新增用戶從一月的約 25 萬成長到三月初的逾百萬。商業成功愈大，生態系統的規則就會愈嚴格。好不好用，試了才知道——但先確認你的使用方式符合 ToS。
🇺🇸 OpenCode Forced to Drop Claude: Anthropic Draws the Line on OAuth Token Abuse
In late March, a popular open-source project with over 125,000 GitHub stars quietly pushed a commit that sent shockwaves through the developer community. OpenCode — a widely-used terminal AI coding assistant — removed all Claude model integrations. The commit message was blunt: anthropic legal requests.
What Is OpenCode?
OpenCode is an open-source command-line AI coding assistant written in Go, comparable to Claude Code or GitHub Copilot but fully self-hostable and community-maintained. It supports multiple AI models, with Claude being particularly favored for its reasoning quality.
The problem: many users were not accessing Claude through paid API keys. Instead, they were using OAuth tokens from their Claude Pro/Max subscriptions — essentially spoofing Claude Code's official client headers to make requests to Anthropic's servers. In practical terms: using a $20/month subscription plan as an unlimited API.
Why Did Anthropic Act?
This was not sudden. The timeline:
- January 9, 2026: Anthropic activated server-side blocks. Third-party tools using OAuth tokens began receiving: "This credential is only authorized for use with Claude Code."
- February 19, 2026: Anthropic formally updated its Terms of Service, explicitly prohibiting OAuth token use from Free/Pro/Max plans with third-party tools or the Agent SDK.
- March 2026: Anthropic issued legal demands. OpenCode pushed a commit removing all Claude integration.
From a business perspective, the logic is clear: users tunneling API traffic through a $20/month subscription were significantly eroding Anthropic's API revenue and adding infrastructure pressure.
Developer Reaction — and the Deeper Problem
The removal commit received 40 thumbs-down and only 4 thumbs-up on GitHub. The HackerNews thread surpassed 398 comments, with opinions sharply divided:
- Anthropic defenders: Subscription terms never permitted this usage. Users were exploiting a loophole — Anthropic is within its rights to enforce its ToS.
- Critics: Claude Code itself is just a CLI tool. By allowing only its own product to use OAuth authentication while blocking open-source alternatives, Anthropic is effectively engaging in anti-competitive behavior within its own ecosystem.
The deeper issue: open-source AI tooling ecosystems remain fundamentally dependent on the goodwill — or tolerance — of commercial model providers. When the same company is both a model provider and a competing tool vendor, ecosystem independence is fragile by design.
Practical Takeaways for Developers
- If your toolchain depends on Claude, use official API keys — not OAuth tokens from subscriptions.
- Adopt a multi-model strategy (OpenAI, Google Gemini, local Ollama) to reduce single-vendor dependency risk.
- Monitor Anthropic's ToS changes closely, especially around Agent SDK usage restrictions.
For context: Anthropic's paid subscriber base has more than doubled this year — daily new signups grew from ~250,000 in January to over 1 million by early March. The bigger a commercial AI ecosystem grows, the stricter its rules become. You won't know until you try it — just make sure your usage is ToS-compliant first.
Sources / 資料來源
- Anthropic takes legal action against OpenCode | Hacker News
- Anthropic clarifies ban on third-party tool access to Claude | The Register
- Anthropic's Claude popularity with paying consumers is skyrocketing | TechCrunch
AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends
留言
張貼留言