MCP & AI
Jasefly is AI-first by design: a Model Context Protocol server exposes controlled cms_* tools so agents can read the site map, update content, run diagnostics, and ship releases — without giving the model shell root on your host.
Repos: clone Core (~15 host modules) for the platform host; domain packages live in Jasefly-Modules (15 first-party modules).
Honest caveat: without MCP configured and a local GitHub clone of the project with MCP wired on both sides, an AI agent does not work with your site directly. Install ZIP alone is not a magic remote AI connection.
- Map — cms_site_map / digests — understand pages, nav, singletons without scraping HTML.
- Edit — cms_update / cms_put_singleton — change layouts and settings with structured JSON.
- Code — Work in the local clone — modules, widgets, PHP — then verify with local tests.
- Release — cms_release — build, package, and deploy the update ZIP to hosting.
What MCP is good for
Controlled operations — not unbounded server access.
- Site intelligence — Pages digest, SEO fields, widget inventory for agents.
- Content ops — Create/update/publish resources with the same admin API.
- Managed deploy — cms_release ships frontend/backend changes as hosting updates.
- Diagnostics — Alive checks, last error, hosting guards before you break production.
- Token-scoped — Operations go through API auth — not SSH as root.
- Framework-native — Tools map to platform concepts: pages, modules, SDK reports.
Get the source and configure MCP
Clone the repo, set mcp-cms credentials locally, then connect your IDE agent.