Cursor has shipped something quietly significant. On April 28, 2026, the company released the Cursor SDK, a TypeScript API that lets developers programmatically create, run, and manage Cursor’s coding agents from their own code, scripts, CI/CD pipelines, or products. It is available in public beta to all users today. The announcement matters because of what it signals about where Cursor is headed, not just what it does.
From Tool to Platform
Until this week, running a Cursor agent meant opening the Cursor desktop app, CLI, or web interface. The SDK removes that constraint entirely. With npm install @cursor/sdk, any developer can spin up a Cursor agent in a few lines of TypeScript, point it at a repository, and have it start working: summarizing code, fixing bugs, opening pull requests, without a human sitting in front of an editor. The SDK exposes the full Cursor harness: codebase indexing, semantic search, MCP server connections, subagent delegation, and support for every model Cursor supports, including its own Composer 2 model. When running on cloud, each agent gets a dedicated virtual machine with strong sandboxing and a cloned development environment. Agents keep running when a laptop sleeps or a network drops, and can reconnect mid-task. This is infrastructure, not a feature.
What Teams Are Already Building
Cursor’s blog post names Faire, Rippling, Notion, and C3 AI as early adopters. The use cases break down into three patterns. First, engineering teams are wiring agents into CI/CD pipelines to automatically identify the root cause of build failures, summarize changes, and push fixes to pull requests without human intervention. Second, companies are building internal tools, platforms where non-engineers like GTM teams can query product data or spin up prototypes without writing code. Third, and most strategically, some companies are embedding Cursor agents directly inside their own customer-facing products, giving end users an AI coding experience without ever leaving the host application. That third use case is the one to watch. It means Cursor’s intelligence is beginning to show up inside products that have nothing to do with Cursor itself.
The Platform Logic
The shift from tool to platform follows a pattern that has played out before in software. Stripe did not just build a payment product. It built a payment API that let other businesses skip the complexity of financial infrastructure. Twilio did the same for communications. In both cases, the companies that embedded those APIs were not going to rebuild them from scratch. The switching cost became structural. Cursor is running the same play. A team that embeds Cursor agents into their CI/CD pipeline or their internal product has wired Cursor into their engineering workflow at a level that is hard to unwind. The SDK makes Cursor less like a product developers choose each morning and more like infrastructure they depend on continuously. This also reframes Cursor’s valuation. The company is reportedly in talks to raise at a $50 billion valuation, up from $2.5 billion just 15 months ago. Priced as an IDE, even a very good one, that number is aggressive. Priced as the agent runtime layer for the software industry, it becomes more legible.
The Broader Context
The SDK launch comes as the AI coding market is intensifying sharply. Anthropic’s Claude Code, OpenAI’s revamped Codex, and GitHub Copilot are all competing for the same developer workflows. Each has a different architectural bet: Copilot lives inside GitHub’s ecosystem, Claude Code leads on context window depth, and Codex runs as a standalone cloud agent. Cursor’s bet is that none of them will be embedded in other companies’ products at scale, because none of them have made the move Cursor just made. The Cursor SDK is available now at cursor.com, billed on token-based consumption pricing.
