Anthropic Buys Stainless SDK Startup

Executive Summary

Fact: Anthropic has acquired Stainless, the New York startup whose technology generates SDKs, CLIs, and MCP servers for OpenAI, Google, Meta’s Llama Stack, Groq, Cerebras, and hundreds of other AI platforms — in a deal reported above $300 million.
Impact: Stainless has immediately shut down its hosted SDK generator. Existing customers keep the code they’ve built, but the tap is off for new users and new SDK creation.
Surprise: Anthropic now owns the company that wrote the developer libraries millions of engineers use to access OpenAI. A key piece of its rival’s developer experience just changed hands.

Most AI acquisitions are about the model. This one is about the pipes.

On Monday, Anthropic confirmed it had acquired Stainless, a New York startup founded in 2022 by former Stripe engineer Alex Rattray. The deal was widely reported above $300 million, backed by investors including Sequoia Capital and Andreessen Horowitz. Stainless does something unglamorous and utterly essential: it takes an API specification and automatically generates software development kits — SDKs — across TypeScript, Python, Go, Java, Kotlin, and more. Those SDKs are how engineers interact with AI APIs in production. Millions of developers download them every week without knowing or caring who built them.

That invisibility is exactly what made this acquisition interesting.

The infrastructure layer nobody was watching

When OpenAI releases a new model, the part that shows up in your Jupyter notebook or your enterprise codebase is not the model itself — it’s an SDK. The Python library. The TypeScript package. The MCP server connector. Stainless generated all of those. At the time of acquisition, OpenAI, Google’s AI platforms, Meta’s Llama Stack, Runway, Groq, Cerebras, LangChain, Braintrust, and Writer all relied on Stainless in their developer toolchain.

That is not a vendor list. That is most of the AI industry’s developer infrastructure sitting inside one small startup. And Anthropic just bought it.

What’s striking about this move is that Anthropic was already a Stainless customer. The Claude SDKs that enterprise developers use to call Claude’s API were built on Stainless tooling. Bringing that in-house is straightforward vertical integration. The more disruptive consequence is what it means for everyone else on that customer list — including Anthropic’s two largest competitors.

A $300M infrastructure grab — and its collateral damage

Stainless announced it will wind down all hosted products immediately. New signups, new projects, new SDK generation: closed. Existing customers retain full rights to the SDKs they’ve built and can modify and extend them freely. But the generator — the engine that keeps those SDKs current as APIs evolve — now belongs to Anthropic.

For OpenAI and Google, this creates an uncomfortable dependency. Their developer-facing libraries were maintained by a company that is now a wholly-owned subsidiary of their closest competitor. They can fork, they can rebuild, but neither is free. Every hour their engineering teams spend on SDK maintenance is an hour not spent on model improvement or product development.

For enterprise buyers — the CTOs and COOs making AI stack decisions — the story lands differently. You probably don’t maintain your own SDKs. You use them. But the acquisition is a reminder that the AI infrastructure layer is not neutral territory. It is contested. As TechCrunch reported, hundreds of companies across the AI ecosystem depended on Stainless-generated code.

Is your AI vendor stack built on infrastructure your providers actually control? Epinium Transform maps the dependency layers in your AI adoption so you’re not surprised when the plumbing changes →

Why MCP servers make this move considerably larger

Stainless didn’t just generate SDKs. It also generated MCP servers — the connectors that let AI agents interact with external APIs at runtime. As agentic use cases move from pilots to production, the MCP layer is becoming load-bearing infrastructure. An enterprise deploying AI agents at scale needs MCP servers that are accurate, consistently maintained, and auditable.

What we’re seeing at Epinium is that MCP governance — not model selection — has become the real bottleneck in most enterprise AI deployments. Anthropic acquiring the tooling that generates MCP servers is not a peripheral move. It is a direct play for the agentic infrastructure layer.

Epinium data

Across 12 brand-side MCP deployments Epinium reviewed in Q4 2025 and Q1 2026, SDK consistency was the most common failure point — surfacing before any model quality issue. Teams that inherited SDKs without understanding their maintenance chain faced the longest remediation cycles when upstream tooling changed. That finding makes the Stainless wind-down immediately relevant for any enterprise running agentic workflows.

For more on what enterprise MCP governance actually requires in production, the MCP Enterprise Security and Governance guide Epinium published this month draws on real deployments, not theory.

Here’s the contrarian read, though: Anthropic may not intend this as a competitive weapon. The company said publicly that existing customers keep their code and their rights. What Anthropic most likely wants is speed — the ability to ship Claude SDK updates faster by owning the tooling internally. That’s a reasonable and unsexy goal.

But intent and consequence are different things. The consequence is that developer infrastructure for AI is consolidating, and Anthropic has just taken a meaningful position at that layer. The question for any enterprise choosing an AI stack in 2026 is not just “which model is best?” It is “who controls the stack below the model?”

FAQ: Anthropic’s Stainless acquisition

Will OpenAI’s Python and TypeScript SDKs stop working?

No. Stainless customers, including OpenAI, retain full rights to the SDKs they’ve already generated. The code is theirs to modify and extend. What they lose is the hosted generator — meaning future API changes require internal engineering effort or a rebuild on alternative tooling, rather than an automated update from Stainless.

Does this give Anthropic access to OpenAI’s private API specifications?

There is no evidence of that. Stainless generated SDKs from specs its customers provided; it was a tooling vendor, not a data custodian. The acquisition moves the tooling team, not the underlying customer specs. That said, Anthropic now employs the engineers who built and maintained code running inside OpenAI’s developer infrastructure — which is significant regardless of data access.

What should enterprise teams do if they’ve built on Stainless-generated SDKs?

First, audit which of your AI vendor SDKs were Stainless-built. For most enterprises using major platforms, the immediate impact is zero — existing code runs as before. The medium-term question is SDK maintenance: if your deployment depends on auto-generated SDK updates tracking a fast-moving API, you’ll need to plan for that capability moving in-house or to an alternative provider before the next major API revision.

Are there viable alternatives to Stainless for SDK generation?

Yes. Speakeasy, Fern, and LibLab offer comparable SDK generation from OpenAPI specs. None currently matches Stainless’s scale or its depth of AI-specific templates built over four years of serving frontier labs. But all three are viable, and all three are likely to see a significant surge in enterprise interest in the coming weeks as Stainless customers plan for the transition.

When does AI stack provenance matter more than model quality?

When you’re building for production at scale, not running demos. At the prototype stage, model quality is the dominant variable. Once you’re deploying agents, integrating into core systems, or managing compliance obligations, the infrastructure layer — SDK maintenance, MCP governance, vendor dependencies, supply chain visibility — starts to matter as much or more. The Stainless acquisition is a clear marker of where that threshold sits in 2026.

The Stainless acquisition is not a model release. No benchmark improved. No demo went viral. But if you’re building anything production-grade with AI this year, the story of who controls the developer tooling layer matters — because that layer is exactly what stands between your business and the model you chose to trust.

Ready to map your AI vendor dependencies before the next infrastructure shift? Epinium Transform helps brand operators and mid-market companies audit and redesign their AI stack — from model selection through to SDK governance and agent deployment. Discover how Epinium Transform builds resilient AI stacks →