Now with AI-powered page building via MCP Server

AI tooling

context7, the MCP server and llms.txt answer three different questions. Which to install, and when.

Three things around cmssy get called "AI support", and they answer three different questions. Getting them straight takes a minute and saves an afternoon.

SurfaceAnswersWho reads it
context7"How do I write this with the cmssy SDK?"Your coding agent, while you build
MCP server"Change this content for me."An agent acting on your workspace
llms.txt"What is cmssy?"Any assistant reading the public web

Only the middle one touches your content. The other two are documentation - the difference between an agent that knows the API and an agent that can use it.

context7: current SDK docs, inside your editor

A model trained months ago will invent a @cmssy/next export that never existed, and it will do it confidently. context7 is the standard fix: an MCP server that fetches a library's current documentation at the moment you ask and drops it into the prompt.

Two cmssy libraries are indexed:

  • /cmssy-io/cmssy-sdk - the headless SDK. Block registration, createCmssyPage, server loaders, draft preview, i18n, the editor bridge.
  • /cmssy-io/cmssy-next-starter - the runnable Next.js starter, as a worked example.

Install it

context7 ships a setup command per client:

npx ctx7 setup --claude   # Claude Code
npx ctx7 setup --cursor   # Cursor

Editors that take a raw MCP config want the hosted server instead:

{
  "mcpServers": {
    "context7": {
      "url": "https://mcp.context7.com/mcp"
    }
  }
}

See context7.com for the current client list and API-key options.

Ask for it by name

Most agents reach for context7 on their own once it is installed. When one doesn't, say so - and naming the library id removes the guesswork entirely:

How do I register a block with a server loader? use context7 /cmssy-io/cmssy-sdk

Worth doing whenever the answer depends on a version. The SDK's block and config APIs have changed shape more than once, and a model without current docs will reproduce whichever shape it was trained on - which compiles, runs, and renders nothing.

llms.txt

cmssy.com/llms.txt is a single page: what cmssy is, the handful of facts that shape everything else - blocks are typed and live in your code, models deliver over a typed GraphQL API, every field is multilingual - and links into the docs and the blog.

It is orientation, not API reference. Point a general-purpose assistant at it when it needs to know what cmssy is. Point a coding agent at context7 when it needs to know how something works.

Which one when

  • Writing block components, routes, loaders → context7.
  • Creating pages, editing copy, importing records → the MCP server, or the assistant inside the admin.
  • Explaining cmssy to a model that has never seen it → llms.txt.

They compose. An agent with both context7 and the cmssy MCP server installed can write a block component and then fill it with real content, in one session - reading the SDK docs for the first half and calling workspace tools for the second.

Next steps