Anthropic legally removed Claude integration from OpenCode, Astral acquired by OpenAI

On March 19, 2026, there were two major movements surrounding AI coding tools. Anthropic took legal action against OSS coding agent “OpenCode” to remove Claude integration, and Python toolchain Astral Software was acquired by OpenAI and joined the Codex team. Coincidentally, they were held on the same day, but the theme of “Distance between AI vendors and OSS” is a continuation of the two.

How Claude integration disappeared from OpenCode

PR #18186 has been merged into OpenCode’s GitHub repository. The title is “Remove anthropic references per legal requests.” Gathered 315 points and 265 comments on HN. At the request of Anthropic’s legal department, all Claude-related features were removed.

Deleted

File	Contents
`anthropic-20250930.txt`	Anthropic branded system prompts
`opencode-anthropic-auth@0.0.13`	Claude Pro/Max OAuth authentication plugin
`claude-code-20250219`	`anthropic-beta`Request header
`providers.ts`	Anthropic tips for provider selection UI
`providers.mdx`	Claude Pro/Max OAuth flow documentation

The biggest impact is the removal of the opencode-anthropic-auth plugin. Claude Pro/Max subscribers could log in to OpenCode using claude.ai’s OAuth flow and use models without an additional API key.

OpenCode’s position

OpenCode (formerly opencode) is an OSS AI coding agent developed by the SST (Serverless Stack) team. The maintainer is thdxr (Dax Raad). He was popular as a rival to Claude Code, Cursor, and Copilot.

With a provider-independent design, you can switch between Anthropic, OpenAI, Gemini, etc. with a plugin. As with the cases of Google and Stripe that I wrote about in Article on Agent Design Principles, the more production operations progress, the stronger the need to not have a fixed model provider. opencode-anthropic-auth had the most users among them.

Why did Anthropic stop?

Although the direct reason has not been made public, it can be inferred from the technical context.

claude.ai’s OAuth flow is designed for use with official clients (claude.ai/Claude Code). If a third-party app uses this flow, problems will arise with the handling of authentication tokens, interpretation of terms of service, and brand usage rights. The removed system prompt anthropic-20250930.txt is branded with Anthropic branding and appears to be an unofficial client using the official system prompt as is.

A similar problem occurs repeatedly with GitHub Copilot. GitHub has tightened restrictions as the community attempts to use Copilot from unofficial clients. The question of which clients can use the subscription-based AI model is a structural one, not specific to Anthropic.

The analysis of Greptile, which automatically executes code reviews, is also interesting. The refactoring of llm.ts due to plug-in deletion introduced a side effect in which the User-Agent: opencode/VERSION header was mistakenly deleted for non-Anthropic providers such as OpenAI, Google, and Azure. There are negative effects of hastily deleting it.

flowchart TD
    A[Claude Pro/Maxサブスクライバー] -->|OAuthフロー| B[claude.ai認証サーバー]
    B -->|認証トークン発行| C[opencode-anthropic-auth プラグイン]
    C --> D[OpenCode経由でモデル呼び出し]
    D --> E[Anthropic API]
    E -.->|法的措置| F[プラグイン削除要求]
    F --> G[PR #18186マージ]
    G --> H[コアプラグイン削除]
    G --> I[コミュニティが<br/>opencode-claude-auth を独自公開]
    I --> D

Community reaction

The PR comment section was a mess. Reactions such as cancel your subscription'' and switch to OpenAI” arose one after another, and some raised the question of @where is the line between legitimate enforcement of product boundaries and hostile moves toward third-party harnesses?”

On the other hand, there were some sober comments. ben-pr-p analyzed that “The plugin is just removed from the built-in list, but works as a third-party plugin.” In fact, griffinmartin published opencode-claude-auth on GitHub/npm within a few hours of the PR merge, and an AUR package for ArchLinux was even created on the same day.

OpenCode’s plug-in architecture was working properly. By removing it from the core, Anthropic’s requirements were formally met, but the design allowed for the community to restore it as an independent plugin.

Astral joins OpenAI’s Codex team

Also on March 19, Astral Software announced its participation in OpenAI. According to founder Charlie Marsh’s official blog, the entire development team will join the Codex team. 357 points in HN.

Astral Products

Astral offers three Python tools. Both are made by Rust and are fast.

Tools	Overview
Ruff	Linter/formatter. It’s tens to hundreds of times faster than Flake8 and Black, and has almost replaced traditional lint tools
uv	Package manager. Integrates pip, pip-tools, venv, and pyenv into one binary. Dependency resolution is dozens of times faster than pip
ty	Type checker. Mypy/Pyright rivalry. Under development

According to Marsh, it has hundreds of millions of downloads per month and “became the foundation for modern Python development.” In fact, it is increasingly being adopted by default in CI environments and VS Code extensions.

Why Codex?

Marsh’s explanation is that it is an extension of the founding mission of “improving programming productivity.” As AI is rapidly changing development methods, we decided that building at the forefront of this would have the greatest impact.

OpenAI’s Codex is being developed as a code generation and execution agent (as mentioned in the Codex Security article, it also takes a unique approach in terms of security). If uv and Ruff are integrated, AI agents will be able to handle dependency resolution, formatting, and type checking autonomously. This comes just after the MCP integration between Codex and Figma was announced in Major update of Copilot, and OpenAI probably aims to thicken the tool chain layer of Codex.

OSS continuity

This is what concerns me the most. OpenAI has stated that it will continue to open source Ruff, uv, and ty, and Marsh has also stated, “We will continue to build in an open environment with the community.” There are currently no plans to change the license or suspend development.

However, what “continuation” means will depend on the future. As core developers focus on Codex within OpenAI, the reliance on external contributors will increase. As I wrote in Chardet Relicense Dispute Article, issues with the governance of OSS projects resurface every time the ownership structure changes. There is also a precedent: HashiCorp Terraform’s BSL migration → OpenTofu fork.

The fact that a toolchain with hundreds of millions of monthly downloads is now part of OpenAI is a significant impact on the Python ecosystem. If ty is completed, it will affect the share of mypy/Pyright, and the tight integration of Codex and uv/Ruff may lead to the formation of a “standard procedure for AI agents to operate Python projects.”

Anthropic was removed through legal action, and OpenAI was acquired through acquisition. It may be a coincidence that two cases of “drawing the boundaries of one’s own ecosystem” with completely opposite approaches occurred on the same day, but I think it means that the area surrounding AI coding tools has entered such a stage.