AI documentation, end to end
AI helps you write it. AI helps you translate it. And now AI can read it — your published docs become a live source that agents and assistants query directly.
“AI documentation” used to mean one thing: using AI to help you write faster. That is still half the story, and OpenDocs does it well — an AI Write Assistant to draft, an AI Writer Improver to tighten, and AI Translations to reach a global audience. But there is a second half that most documentation tools have not caught up to yet. AI assistants now consume documentation. When a developer asks their coding agent how your API works, that agent needs a clean, current, machine-readable source — not a scraped web page or a stale export.
OpenDocs treats both halves as first-class. You author with AI, you localize with AI, and your published content doubles as a queryable knowledge source for AI agents through a built-in MCP server. The rest of this page walks through all three pillars, why the BYOK model matters for the AI features, and where the AI-readable docs ecosystem is heading.
Write with AI
AI Write Assistant
Start from a prompt instead of a blank page. The AI Write Assistant drafts sections, outlines, and first passes directly inside the block editor, so you spend your time refining rather than staring at an empty screen. It runs on every plan.
AI Writer Improver
Tighten what you already have. The AI Writer Improver rewrites for clarity, trims filler, and smooths tone — useful when a rough internal note needs to become a polished public page. You stay in control of what you accept.
BYOK, no markup
Both writing tools are bring-your-own-key. You connect your own Anthropic API key, so the AI runs on your account and your data relationship — and usage is billed by Anthropic at their rates, never marked up by OpenDocs.
AI Write Assistant
Draft sections, outlines, and full first passes with Claude, Anthropic's AI, without leaving the block editor. Paste your own Anthropic API key and it runs on every plan — on your account, under your data relationship, at Anthropic's own rates.
The writing tools live where you already work: the OpenDocs block editor. There is no separate AI console to switch to and no Markdown or git knowledge required to use them. A non-technical writer can draft a page with the Write Assistant, run the Writer Improver over a clumsy paragraph, and publish — all in the same visual editor. Because the AI features are BYOK, the same key powers writing, improving, and translation, and you keep a single, transparent view of what the AI is doing on your Anthropic account.
Translate with AI
38 languages
AI Translations render your documentation in up to 38 languages. Reach a global audience without hiring a localization vendor or standing up a separate translation pipeline for every locale you want to serve.
Stays in sync
Translations are not a one-time dump. When you update the source page, the translated versions stay in sync with your changes, so your Spanish, Japanese, or German readers are not left on an outdated copy.
Global docs, your key
Like the writing tools, translation is BYOK on your Anthropic key. You get multilingual documentation with usage billed straight to your account — predictable, transparent, and without a per-language subscription add-on.
Localization is usually the point where documentation projects stall. Vendors are expensive, freelance translators are slow, and keeping dozens of translated pages current as the product changes is a job in itself. AI Translations collapse that overhead: you publish once, translate into the languages your audience speaks, and let the translations track the source as it evolves. For a small team serving customers in multiple regions, this is often the difference between having localized docs and not having them at all.
Serve your docs to AI
A live MCP server
Every published space is reachable through the OpenDocs MCP server over streamable HTTP. Agents connect to a live endpoint and read your current published content on demand — not a scraped page and not a stale export you have to remember to regenerate.
Four purpose-built tools
The MCP server exposes list_spaces, get_page_tree, get_page (full page content as clean plain text), and search_pages. An agent can enumerate what exists, walk the structure, read a page, and search across pages.
Works with any MCP client
Because it speaks the Model Context Protocol, it works with any MCP-compatible client — Claude Desktop, Claude Code, and other MCP clients. Point the client at your endpoint, authenticate with an API key, and your docs are in the loop.
This is the pillar that changes what documentation is for. Traditionally, docs are written for humans reading a browser. But increasingly the reader is an AI agent acting on a person's behalf — a support assistant answering a customer, a coding agent wiring up your API, an internal bot fielding questions from your team. Those agents work best with a structured, authoritative source they can query, and that is exactly what the OpenDocs MCP server provides.
The endpoint is secured with an OpenDocs API key, so access is controlled rather than wide open. An agent reaches only the content the key is permitted to see, and you can issue or revoke that key. The result is that your published documentation becomes a live, queryable knowledge base for AI — without you exporting files, maintaining a separate vector database, or letting bots scrape your site and drift out of date.
Why BYOK matters
Every AI feature in OpenDocs — the Write Assistant, the Writer Improver, and AI Translations — is bring-your-own-key. You supply your own Anthropic API key, and that single decision has three consequences worth understanding.
Cost transparency. Your AI usage is billed by Anthropic directly to your account, at Anthropic's published rates. You can see, to the token, what generating and translating your docs actually costs, because it shows up on your own invoice rather than being bundled into an opaque platform fee.
No AI markup. Many tools resell AI capacity with a margin baked in, or gate the good models behind a higher tier. OpenDocs does neither. Because you bring your own key, OpenDocs is not in the business of reselling tokens to you — there is nothing to mark up.
You control the key. The data relationship for AI usage is between you and Anthropic, governed by your own account terms. You choose the key, you set its limits, and you can rotate or revoke it. Your AI provider relationship stays yours, not intermediated by your docs vendor.
The AI-readable docs landscape
Making documentation legible to AI is a fast-moving space, and two approaches are emerging. It helps to understand both, because they solve different problems.
llms.txt is a community convention: a plain-text file at a known location that points AI systems to the LLM-friendly parts of a site, a bit like robots.txt or a sitemap but aimed at language models. It is a lightweight, static way to advertise content. It is still emerging, and adoption across tools and crawlers is uneven. To be clear, OpenDocs does not currently generate llms.txt files for published sites — we mention it because it is a real part of the ecosystem and worth watching, not because OpenDocs ships it today.
MCP takes the interactive path. Instead of a static file describing content, the Model Context Protocol gives agents live tools to query content on demand — list what exists, walk the structure, read a page, run a search. It is the difference between handing an agent a map and giving it a phone line to call. For documentation that changes, the live approach means an agent always sees the current state, never a snapshot.
OpenDocs ships MCP today. Your published spaces are queryable right now by any MCP-compatible client, secured with your API key. If llms.txt or similar conventions mature into something broadly useful, they are complementary to what MCP already does — but the interactive, always-current path is the one that matters most for docs that agents rely on, and it is the one OpenDocs provides now.
Traditional docs vs AI-ready docs
| Dimension | Traditional docs | AI-ready docs on OpenDocs |
|---|---|---|
| Primary reader | Humans in a browser | Humans and AI agents |
| How agents access content | Scraping HTML, best effort | Live MCP server, structured tools |
| Freshness for agents | Stale exports or cached crawls | Current published content on demand |
| Authoring assistance | Manual, or a separate AI tool | Built-in Write Assistant & Writer Improver |
| Localization | Vendor or manual pipeline | AI Translations, 38 languages, stays in sync |
| AI cost model | Bundled or marked-up tokens | BYOK — billed by Anthropic, no markup |
| Access control for agents | All or nothing | API-key-secured MCP endpoint |
Frequently asked questions
What is AI documentation?
AI documentation covers two related ideas: using AI to help produce documentation, and making documentation readable by AI. OpenDocs does both. AI helps you write and translate pages, and once published, your docs are exposed to AI agents through a live MCP server so assistants can read and search them directly instead of scraping a website.
Do I need my own API key to use the AI features?
Yes. The AI Write Assistant, AI Writer Improver, and AI Translations are BYOK (bring your own key): you connect your own Anthropic API key. That means the AI relationship, the data terms, and the usage billing are all directly between you and Anthropic. OpenDocs never marks up your AI usage.
What does the AI usage cost?
Because the AI features are BYOK, AI usage is billed by Anthropic to your own account at Anthropic's published rates. OpenDocs does not add a per-token markup or resell tokens. Your OpenDocs subscription covers the platform; your Anthropic account covers the AI usage, so you see exactly what the AI costs.
What is MCP?
MCP (Model Context Protocol) is an open standard that lets AI assistants connect to external tools and data sources. OpenDocs runs an MCP server over streamable HTTP that exposes your published spaces to any MCP-compatible client, such as Claude Desktop or Claude Code. Its tools let an agent list spaces, browse the page tree, read a page as plain text, and search across pages.
Can AI agents read my private docs?
Only with access you grant. The OpenDocs MCP endpoint is secured with an OpenDocs API key, so an agent can only reach the content that key is allowed to see. Access is controlled, not open to the public internet, and you can issue and revoke the key that agents use.
Does OpenDocs generate an llms.txt file?
Not today. llms.txt is an emerging community convention for advertising LLM-friendly content, and it is worth watching, but OpenDocs does not currently generate llms.txt files for published sites. What OpenDocs ships now is the interactive path: a live MCP server that lets agents query your real published content on demand.