OpenCode, the open-source coding agent, shipped a way to run Claude through a user’s own Claude Code login on October 30, 2026. Version 2.2.3, released at 09:55 UTC, adds a Claude Agent provider: install Claude Code, sign in with it, then pick Claude Agent in OpenCode’s connect menu. Later that day the project posted on X: “you can now use your claude subscription in OpenCode”.
That is the part people noticed, because Anthropic shut this door in January. It said then that third-party harnesses using Claude subscriptions break its terms, and its Agent SDK documentation still says: “Unless previously approved, Anthropic does not allow third party developers to offer claude.ai login or rate limits for their products, including agents built on the Claude Agent SDK.” Anthropic has said nothing public about OpenCode’s new provider. The provider’s author says Anthropic gave permission, as the correction below sets out.
Correction, November 14, 2026: this page said we found no sign that Anthropic had approved the provider. That was wrong. On October 30 at 16:24 UTC, before this page was published, the author of the pull request wrote on it: “We got permission from Anthropic and everything here is fully tos compliant”. GitHub marks him as a member of the organization that owns the OpenCode repository. We missed the comment. Anthropic has not confirmed it in public, and the comment does not say who gave the permission or what it covers. The answers, the table and the picture below are updated, and a new section lists what has changed since October 30.
On this page · 6 sectionsOpen
OpenCode 2.2.3, released October 30, 2026, adds a Claude Agent provider that runs Claude through the user’s own Claude Code login.
It runs the user’s installed claude program through Anthropic’s Agent SDK and never reads or replays the saved login, per its spec.
A subscription, an API key or a cloud account all work; on a subscription the cost display reads zero.
Anthropic’s Agent SDK docs bar offering claude.ai login in third-party products unless previously approved.
Anthropic banned subscription use in third-party harnesses in January 2026. The provider’s author says Anthropic gave permission for this one; Anthropic had not confirmed that in public by November 14.
§ 01How the new provider works
The design is in the provider spec linked from the pull request, opened October 24 and merged October 30. OpenCode does not copy or replay a Claude login. It runs the user’s installed claude program through Anthropic’s own Claude Agent SDK. The spec puts it plainly: “It never sits between Claude Code and Anthropic or changes Claude Code’s requests.”
| Part | What happens |
|---|---|
| Login | The user signs in with Claude Code in a terminal; OpenCode never reads the saved login |
| Logins accepted | A subscription, an API key or a cloud account |
| System prompt | Claude Code’s own prompt, plus a short note that it runs in OpenCode |
| Tools | Claude Code’s built-in tools are off; Claude gets OpenCode’s tools instead |
| Cost shown | Zero on a subscription; the model’s list price on an API key or cloud account |
On a subscription, the spec says, the cost display reads zero: “A subscription shows $0 because it covers usage.”
§ 02Why January was different
On January 9, 2026 Thariq Shihipar of Anthropic’s Claude Code team explained a crackdown from the day before: “Yesterday we tightened our safeguards against spoofing the Claude Code harness after accounts were banned for triggering abuse filters from third-party harnesses using Claude subscriptions.” His next post set the rule: “Third-party harnesses using Claude subscriptions create problems for users and are prohibited by our Terms of Service.” And the alternative: “This is why the supported way to use Claude in your own tools is via the API.”
The tools caught in January pretended to be Claude Code. OpenCode’s new provider runs the real Claude Code program, with its own prompt and telemetry, and only swaps in OpenCode’s tools and interface. That answers the technical complaint Anthropic made then, unusual traffic without Claude Code’s telemetry. It does not answer the policy one. The SDK note bars offering claude.ai login in a third-party product unless Anthropic approved it, and the spec’s own open questions show the authors were not sure when they wrote it: “Do we need to discuss anything here with anthropic or any kind of approval?” On October 30 the pull request’s author wrote that Anthropic had given permission.
§ 03What is confirmed, and what is not
| Claim | Status |
|---|---|
| OpenCode can run Claude through a Claude Code login | Confirmed: release 2.2.3 and the merged pull request |
| A subscription login works | Stated by OpenCode and its spec |
| It uses Anthropic’s public Agent SDK | Confirmed in the spec |
| Anthropic approved it | Stated by the provider’s author on the pull request; not confirmed by Anthropic in public |
| Anthropic’s docs allow it without approval | No: the SDK note bars claude.ai login unless approved |
| Accounts risk a ban | Unknown; January’s bans hit tools that spoofed Claude Code |
§ 04What to do if you use OpenCode with Claude
- An API key is the safe path. It is the method Anthropic’s SDK docs point to, and the provider accepts it. On a key, OpenCode shows the model’s list price.
- A subscription login rests on the author’s word. He says Anthropic gave permission. Anthropic has not said so in public, and in January accounts were banned and later reinstated.
- Watch for an Anthropic statement. A post from the Claude Code team, a change to the SDK note, or a new consent screen would settle it.
§ 05What changed since October 30
- The author says Anthropic gave permission. Asked on the pull request whether the approach had been checked with Anthropic, its author answered on October 30 at 16:24 UTC: “We got permission from Anthropic and everything here is fully tos compliant”.
- Nobody outside the two companies has seen the permission. Later commenters on the pull request asked who gave it and whether there is proof. The author had not answered them when we read the thread at 00:11 UTC on November 14.
- The SDK note is unchanged. At 00:14 UTC on November 14 the Agent SDK page still carried the sentence quoted at the top of this page, with its words “Unless previously approved”.
- Anthropic’s legal page for Claude Code has two sentences that matter. Its legal and compliance page says: “Anthropic does not permit third-party developers to offer Claude.ai login into their own applications, or to route requests through Free, Pro, Max, or Ultra plan credentials on behalf of their users.” It also says its rules do not prevent “an end user from signing in to the unmodified Claude Code binary with their own Claude subscription”. The page carries no date, so we do not know whether either sentence was there on October 30. Anthropic has not said which one applies to OpenCode’s provider.
- The Usage Policy names the practice, not the tool. Anthropic’s Usage Policy, marked effective November 12, 2026, tells users not to: “Resell, proxy, or otherwise provide access to Claude through unauthorized means, including services that route requests through consumer subscriptions or misrepresent the product or client being used”. It does not name OpenCode. Whether a provider that its author says is permitted counts as unauthorized is for Anthropic to say.
- The provider is still in OpenCode. Releases 2.2.7 on November 10 and 2.3.0 on November 13 both carry fixes to it.
- What we searched. A search of X at 00:07 UTC on November 14 covered the past seven days of posts from seven accounts: Anthropic’s, two of its Claude Code staff, and OpenCode’s. It returned one post, on another subject. A web search the same hour found no statement from Anthropic and no report of an account banned for using the provider. That is not proof that neither exists.
§ 06Where CellCog stands
Our conflict, declared: CellCog’s AI employees run on Anthropic’s Claude models (Claude Opus 5.5 on most tiers; Claude Fable 5.2 on Agent Max and Creative Max since November 11, 2026) through Anthropic’s API, not through a consumer subscription, so no subscription rule applies to them. They also work on their own machine with their own accounts, so a dispute over one tool’s login does not stop the work.
Q1How do I set up Claude in OpenCode?
Install Claude Code, sign in with it in a terminal, then choose Claude Agent in OpenCode’s connect menu. Version 2.2.4 added a guided setup.
Q2Can I use an API key instead of a subscription?
Yes. The provider accepts an API key or a cloud account, and Anthropic’s docs name the API key as the supported method.
Q3Could my Claude account be banned?
Anthropic has not said. The provider’s author says it has Anthropic’s permission and follows its terms. In January 2026 accounts were banned for triggering abuse filters through third-party harnesses, then reinstated.
Q4Does OpenCode see my Claude login?
Its spec says no: users sign in with Claude Code, and OpenCode only records which login method they chose.
Q5Does this affect CellCog?
No. CellCog’s AI employees run on Anthropic’s Claude models (Claude Opus 5.5 on most tiers; Claude Fable 5.2 on Agent Max and Creative Max since November 11, 2026) through Anthropic’s API, not a consumer subscription.
