Claude meets Foundation CMS in two places. Claude's search and answer crawlers are allowed on Foundation CMS sites while its training crawler is refused, so sites can be cited without donating their copy. And the content API hands an AI the real schema and the generated contract, so Claude can build against a site's actual content model rather than a pasted copy of it.
Overview
This integration connects Foundation CMS with Claude, Anthropic's assistant. Foundation CMS is Swarm Labs' multi-tenant content management system on Cloudflare Workers. It holds a site's content and hands it to the site as whole, typed records, and every site runs as its own Worker with its own database and media bucket. Anthropic documents three agents: ClaudeBot collects pages for training, Claude-SearchBot indexes them so Claude can cite them, and Claude-User fetches a page on behalf of someone asking about it. The Foundation CMS crawler register names all three and treats them by purpose.
Business Context and Core Use Case
There are two jobs here. The first is visibility: being cited when someone asks Claude a question the site answers. The second is building: agencies increasingly use AI to write front ends and automations, and an AI that guesses at a CMS's fields produces code that compiles and then breaks. The Foundation CMS content API exists partly so an AI can fetch the real model and contract first.
The Applications Involved
Foundation CMS (Foundation CMS) writes the crawler rules and serves the schema, contract and content through its API to a key-holder.
Claude (Claude) cites sites through its search crawler and, as a coding assistant, builds against what the API returns.
How the Integration Works
For visibility, each site's robots.txt allows Claude-SearchBot and Claude-User and refuses ClaudeBot by default, with a content signal line stating that search and AI answers are welcome and training is not. For building, a site-scoped key lets Claude read /api/v1/schema for the applied model, /api/v1/contract for the generated TypeScript types and /api/v1/help for the endpoints that key may call, then write code that matches the site exactly.
Immediate Operational Value
Sites can be cited by Claude without opting into training, and AI-written integrations stop drifting from the real model: if a field is missing, the answer is a change in the CMS that regenerates the contract, never a hand-edited type.
Security, Access, and Governance
API keys are scoped to one site and named abilities, hashed at rest and shown once. A key for building can be read-only, so an assistant can study the model without being able to change content. Crawler choices are per site and set by administrators.
Summary
Claude can cite Foundation CMS sites without training on them, and build against their real content model through the API, so what it writes fits the site it is for.
Frequently asked questions
Can Claude cite a Foundation CMS site?
Yes. Claude-SearchBot and Claude-User are allowed by default, so Claude can index and fetch the site's pages.
Is the content used to train Anthropic's models?
Not by default. ClaudeBot, Anthropic's training crawler, is refused unless the administrator allows it.
Can Claude build a site against Foundation CMS?
Yes. With a site-scoped key it can read the schema and the generated TypeScript contract, so its code matches the real content model.
Does a build key let the AI change content?
Only if it was given write abilities. A read-only key lets it study the model and content without changing anything.
