A year ago, putting an "Open in ChatGPT" button on your API docs felt novel. Today, it's expected. The question for API platforms isn't whether to make documentation AI-accessible; it's about the accuracy and depth AI has about your API.
This post covers our recent improvements to llms.txt, introduces the new AI Context Menu, and then makes the case that even with both, the real differentiator for API developer onboarding is something deeper: Context Plugins.
What we improved in llms.txt
llms.txt — the emerging convention for surfacing structured, AI-readable summaries of a site's content — has quietly become one of the most important developer-experience standards in years. Done well, it lets any LLM understand your API without scraping fragile HTML.
In this release, we revamped our llms.txt generation and shipped per-page Markdown exports alongside it:
- Per-page Markdown exports — every page in your portal is now available as its own Markdown document, giving AI tools a clean page-level source instead of the full site
- A restructured
llms.txtandllms-full.txtthat mirror your portal's navigation and link directly to page-level Markdown, so AI tools can follow the docs the way users actually explore them - Far richer
llms-full.txtcontent: authentication flows, request and response details, errors, usage examples, server info, model descriptions,OneOfandAnyOfinitialization examples, and rendered guide content like tabs, callouts, and code blocks
The result: ask Claude or ChatGPT a question about your API, and the answer is grounded in current, structured, accurate context — not a hallucinated guess assembled from training data.
The AI Context Menu is table stakes now
Alongside the llms.txt work, we shipped the AI Context Menu — the one-click handoff that sends any API page directly into ChatGPT, Claude, Perplexity, Grok, or Google AI Studio, with the full page as structured context. Or copy it as clean markdown for any other AI assistant.
It's a great feature. It's also increasingly the minimum.
Most documentation platforms will have something like it by the end of the year, if they don't already. And rightly so: it removes friction from the moment a developer says "let me ask my AI about this" — which, let's be honest, is most moments.
But here's the limitation of any AI Context Menu, no matter how well it's built:
It optimizes for asking questions. It doesn't optimize for shipping code.
A developer using the AI Context Menu to ask "how does this endpoint work?" gets a fast, accurate answer. The same developer, sitting down to build a production integration, needs much more than answers. They need their AI assistant to understand their codebase, their auth flow, their error-handling conventions, their deployment target — and to generate API integration code that follows all of it.
That's where Context Plugins come in.
Context Plugins: built for production integrations
Context Plugins are the deeper layer of AI-readiness. Where llms.txt and the AI Context Menu makes your API understandable to an LLM in a chat window, Context Plugins make your API understandable inside the developer's own AI tooling — Cursor, Claude Code, Copilot, whatever they use to actually write code.
A Context Plugin gives the developer's AI:
- The full SDK surface for the language they're working in, not just the REST contract
- Idiomatic code examples that match the SDK's conventions, not generic curl snippets
- Auth setup, configuration, and error-handling patterns as part of the context, not as an afterthought
- Version pinning, so the AI generates code that compiles against the version the developer is actually using
The difference shows up in what the developer ships.
With the AI Context Menu, a developer asks "how do I create a customer?" and gets a correct answer — usually a code snippet in whatever language the model defaults to. They then translate it into their actual codebase, retrofit it to their conventions, and debug the parts that didn't quite line up.
With a Context Plugin installed, the developer asks their AI inside their editor to "add a customer creation flow to the signup handler" — and gets code that uses the right SDK methods, the team's existing error-handling utility, the configured client from DI, and the right test patterns. Because the AI has the full picture.
It's the difference between a smart consultant on a Zoom call and a teammate who already has commit access.
Why this matters for onboarding
The highest cost in API adoption isn't the first successful request. It's everything between that request and a production-ready integration: handling retries, threading auth through middleware, pagination, observability, idempotency, and edge cases.
The AI Context Menu shortens the path to the first request. Context Plugins shortens the path to production-ready integration.
That's why we've invested so heavily in them. For the API providers using our API Experience Platform, this changes what their consumers can ship — and how fast. A weekend prototype becomes a Monday deploy.
What to use when
To be clear, this isn't an either/or. Both should exist in any mature API experience:
llms.txt+ AI Context Menu: for exploration, debugging, ad-hoc questions, and developers who are still evaluating the API- Context Plugins: for the actual work of building, integrating, and shipping
If you're an API provider, the takeaway is simple: don't stop at the menu. The AI Context Menu is the start of an AI-native developer experience, not the finish line.
Resources
- Context Plugins: an overview — what they are and how they work
- Case study — see Context Plugins in action
- Use cases — A few vibe-coded applications with a single prompt
