Anthropic's Claude Code Integration: A Legal Minefield

There's a phrase I've heard roughly ten thousand times in my career: "Just ask for forgiveness, not permission." Usually it's code for "I don't want to wait for legal to get back to me." But Theo, the developer behind T3 Code, is currently living in a space where he can't ask for permission because nobody will answer, and forgiveness might come in the form of user account bans.

T3 Code now supports Claude Code subscriptions. This should be uncontroversial—it's just another UI option for people who've already paid for the service. Except Anthropic has been sending lawyers after other open-source projects that tried similar integrations, and nobody seems to know exactly where the line is.

The Open Code Problem

Here's what happened to Open Code, another open-source coding tool: They integrated Claude support. It worked great. Then Anthropic's lawyers showed up. Open Code had to pull their Claude Max plugin, announcing: "We did our best to convince Anthropic to support developer choice, but they sent lawyers instead."

Anthropics's objection, as far as anyone can tell, is about how the integration works. Open Code built their own "harness"—the underlying system that manages authentication and API calls. Anthropic's terms of service say third-party developers can't offer Claude.AI login or rate limits through their products. Open Code had to create their own OAuth flow, fake headers, manage subscription tokens. From Anthropic's perspective, that crossed a line.

T3 Code took a different approach. Theo's team uses Anthropic's official SDK and CLI instead of building their own infrastructure. They're not managing authentication—they're just calling your locally installed Claude Code CLI. It's the difference between picking someone's lock versus asking them to unlock the door for you.

Is that distinction meaningful? Theo thinks so. He even asked Claude itself to review the implementation against Anthropic's terms of service. Claude agreed they're compliant. Whether that would hold up if actual Anthropic lawyers got involved is a different question entirely.

The $5,000 Subsidy Nobody Talks About

Here's the part that makes all of this particularly thorny: Anthropic's Claude Code subscription is massively subsidized. You pay $200 a month. You can use up to $5,000 worth of compute.

That's not a typo. That's not a promotional rate. That's their business model.

Why would they do this? "They want to crush the competition," Theo explains. "They want everyone to have these high tier subs and just use anthropic models. And we can't compete with that. Nobody can compete with that."

This creates a brutal dynamic for anyone trying to build competing services. Users get accustomed to that level of compute for $200. If they switch to a different service that can't afford the same subsidization, they're suddenly getting one-tenth to one-twenty-fifth the inference for the same price. They leave. Anthropic wins.

This also explains why Anthropic might be particularly aggressive about controlling how people use these subscriptions. If third-party tools make it easier to burn through that compute subsidy inefficiently, Anthropic's economics get even worse.

The Silence That Says Everything

What's remarkable about this situation isn't the legal ambiguity—that's standard in emerging tech. What's remarkable is Anthropic's sustained refusal to clarify anything.

Matt Pocock, a prominent TypeScript educator, tried asking straightforward questions: Can he use an OAuth token from a subscription to power the Claude agent SDK locally? If he builds an open-source tool using this pattern, can he distribute it? These aren't trick questions. They're the kind of thing any developer working with an API needs to know.

Anthropic's response, through their DevRel lead: "Sorry, this has been confusing. I know we should be clearer here... We're working on more clarity for this."

That was over a week ago. Radio silence since.

When Pocock followed up with a gentle "ETA?", Theo's reaction captured the frustration: "If someone did this to me on a thing that had been stressing me out as much as I am certain this Claude code drama has been stressing out poor Thoric, I would quit on the spot."

The excuse offered—rapid growth making it hard to scale the team—rings hollow when you're a multi-billion dollar AI company. Especially when the consequence of not clarifying is that developers can't ship products, users don't know if they'll get banned, and everyone operates in a fog of legal uncertainty.

The Pattern We've Seen Before

I've watched this movie before. Company offers generous developer access. Ecosystem starts building. Company realizes the economics don't work or wants more control. Terms tighten. Developers who built in good faith get burned.

What makes this iteration particularly frustrating is the vagueness. Anthropic has already hard-coded blocks—if "open code" appears in API headers, requests get denied for OAuth through subscriptions. That's a clear line. But they won't draw the other lines clearly enough for developers to know where they stand.

Theo's solution—using the official SDK and CLI—seems designed to thread this needle as carefully as possible. "We specifically built T3 code in such a way to make it as likely as possible that we are complicit with these subsidized token options," he explains. They abstracted their event system, avoided building their own harness, did extra work to integrate properly with existing CLIs.

It still might not be enough. As Theo acknowledges: "At any point this goalpost can move because Anthropic loves to do that. They're keeping things vague so they can change their mind at any point."

What Developers Actually Need

This isn't about developers expecting special treatment from a billion-dollar company. It's about needing basic information to do their jobs. Can they build this tool? Will their users get banned? What are the actual rules?

Theo puts it plainly: "We need to be able to talk about these things. We need to be able to release the things we're working on. And we're all stuck in this weird uncertain space that I've never experienced anything like before."

He's not alone in that assessment. Multiple developers have been DMing him asking if he has insights into what's actually allowed, because Anthropic won't say. Even prominent community members who've tried engaging in good faith can't get straight answers.

The irony: Anthropic's DevRel acknowledges the problem. "We've not done fully right by Agent SDK users. I agree, but it's not for lack of trying or wanting." But acknowledgment without action is just sophisticated stalling.

Theo's shipped T3 Code's Claude integration anyway. He's relatively confident it's compliant. He's also explicitly clear this isn't legal advice and he can't control what Anthropic does. If users get banned, he wants to know so he can adjust. If Anthropic goes after accounts using T3 Code specifically, he's offered to cover subscription costs in exchange for permission to name names when he calls them out.

That's not the position any developer should have to take—shipping a feature while offering to compensate users if the vendor bans them for using it. But when the alternative is waiting indefinitely for clarity that may never come, what else do you do?

—Mike Sullivan