Anthropic Buys Its Rivals' SDK Supplier

Anthropic announced this morning that it has acquired Stainless, a four-year-old New York infrastructure startup that has, until today, generated the official SDKs for Anthropic, OpenAI, Google, and Cloudflare. The reported price is north of $300 million. Anthropic will wind down Stainless's hosted products. Existing customers keep the SDKs they have already generated; future generation through the hosted platform goes away.

This is a slightly unusual deal and I want to walk through why it is that way.

What the primary documents say. Anthropic's announcement frames the acquisition around the Model Context Protocol — the open standard Anthropic published in November 2024 for connecting agents to external tools and data. Stainless built MCP server tooling alongside its SDK generation business. Anthropic's note says it is bringing Stainless in to accelerate "the developer infrastructure for agent-era applications" and that hosted products will be wound down. Stainless's own blog post, titled with the genre-standard "Stainless is joining Anthropic", confirms the wind-down and tells existing customers their generated SDKs continue to work but new hosted generation does not.¹²

That second sentence is the deal. Existing artefacts persist; the factory closes.

What Anthropic actually bought. Three things, in descending order of how often they will be discussed:

First, the SDK generation pipeline that produces every official Anthropic client library. This is a vertical-integration move of the most ordinary kind: a platform company internalising a supplier it was already paying. If that were the whole story, the price would be lower and the press release would be shorter.

Second, the MCP server tooling. Anthropic owns the protocol; now it owns the reference implementation infrastructure. This is the part of the deal that matters structurally, and I will come back to it.

Third, and this is where the deal gets interesting, Anthropic bought the fact that OpenAI, Google, and Cloudflare were using Stainless. That is not an asset on a balance sheet. It is a competitive position. Anthropic is paying to remove a shared utility from the market.

The price is a tell. Stainless's revenue was distributed across the major frontier labs and a long tail of API companies. As a standalone SaaS business, $300M-plus is a generous multiple on what was almost certainly a small ARR base for a developer-tools company at this stage. The premium is not on Stainless's intrinsic cash flows. It is on two things the cash flows do not capture: option value on MCP tooling becoming the connective tissue of agent deployments, and the negative option value to rivals of having their SDK pipeline shut down.

Defensive premiums are not new. What is mildly notable is that Anthropic is paying one for developer infrastructure rather than for weights, talent, or compute. That is a statement about where the company thinks the binding constraint sits in the agent era.

Testing the model weight lineage frame. The frame says: in AI, the IP that matters is increasingly embedded in artefacts that don't fit the patent-and-contract model — weights, fine-tunes, agent scaffolds, and now, on this evidence, SDK generation logic and MCP server implementations. The Stainless deal fits the frame cleanly. Anthropic is not licensing the tooling, not partnering, not contributing to an open-source equivalent. It is buying the company and shutting the hosted product. That is the behaviour of an acquirer who believes the IP is non-fungible and that letting rivals continue to access it has a real cost.

The frame holds. It also makes a sharper prediction: if developer-tooling IP genuinely behaves like weight IP, expect more of these acquisitions. The eval harness vendors, the agent-orchestration frameworks, the prompt-management platforms, the MCP tooling alternatives that will now spring up — these are the next candidates. I would watch the cap tables of the better-funded developer-tooling startups with this in mind.

Where the frame strains. MCP is an open protocol. Anthropic published the spec, the community has built against it, and nothing in the Stainless acquisition closes the protocol. Competing MCP server tooling can be built, will be built, by anyone with the engineering capacity. OpenAI and Google have that capacity in surplus. Cloudflare has been building developer infrastructure as a first-class business for over a decade. The migration burden is real and annoying; it is not crippling.

So the moat is narrower than the press cycle will make it sound. What Anthropic has bought is a head start and the right to set the implementation pace, not exclusivity. The frame predicts durable advantage; the open-protocol structure caps how durable that advantage can be.

This is the tension worth holding. The deal logic only works if Anthropic believes (a) MCP becomes the dominant agent connectivity standard and (b) being the best implementer of MCP tooling produces compounding advantages — better Claude-agent integrations, faster iteration on the protocol, preferred-developer status — even when the protocol itself remains nominally open.

What this is a case of. Three patterns at once:

A platform company internalising a shared supplier. Standard playbook, executed at a price that reflects the agent-era stakes rather than the supplier's standalone economics.

A protocol owner buying the reference implementation. Closer to Microsoft acquiring a major Linux distribution than to a typical infrastructure roll-up. The protocol stays open; the best tools become proprietary; "open" gradually means something different in practice than in spec.

An auction outcome, probably. Sequoia and a16z were on the cap table. Stainless's customer list was the customer list. It is hard to imagine OpenAI and Google did not at least look. If Anthropic won an auction, the $300M is partly an auction price, not purely a strategic valuation. I would not over-read the multiple as Anthropic's private estimate of intrinsic value.

What to watch.

The wind-down timeline. Anthropic's announcement is vague on dates. How long OpenAI, Google, and Cloudflare have before hosted generation stops will tell you how aggressively Anthropic is willing to pressure rivals versus how much it wants to avoid antitrust attention.

The next MCP tooling raise. A well-funded open-source-flavoured alternative to Stainless's MCP server tooling is now a near-certain venture bet. Watch for one to be announced within the quarter.

Anthropic's developer platform release cadence. If the Stainless team's output shows up in Claude's developer experience faster than the integration timeline would suggest, the acquisition was about capability. If it doesn't, the acquisition was primarily about denial. Both are defensible; they imply different things about Anthropic's strategic confidence.

The price was the easy part. What it bought is the harder question, and the answer arrives in shipping cadence over the next two quarters.

Footnotes

¶