AI agents

editsy meets AI agents from two directions: coding agents that build sites people can then edit, and content agents that make the edits themselves over MCP.

Agents that build sites

If a coding agent builds or maintains your site, point it at the editsy conventions from your project's agent instructions (AGENTS.md, CLAUDE.md, or whatever your tool reads). npx editsy init sets this up for you: alongside the config and route files it scaffolds an AGENTS.md carrying the conventions, and if you already have one it prints a short section to paste in rather than touching your file. An agent that follows them keeps every visible string in content files, so the site it hands over is editable from the first commit. Add npx editsy check to CI and the constraint holds for every change after that, agent-written or not.

Agents that edit content

@editsy/mcp is an MCP server that gives agents the same editing pipeline the editor UI uses. Four tools: list_content_files, read_content (a file as structured values, plus a field map and a revision id), write_content (values in, a reviewable diff out), and check_content (the same validation as npx editsy check).

The point is the guarantees around the edit rather than the access itself: writes go through editsy's rewriter, so comments and formatting survive, a file that changed underneath the agent is refused instead of overwritten, and nothing outside the content globs can be touched. It runs against the local working tree only, with git as the review step, and no auth or publishing of its own.

Register it from your site's repo. Claude Code:

claude mcp add editsy -- npx -y @editsy/mcp

Clients that take a JSON config (Cursor, Claude Desktop, and friends):

{
  "mcpServers": {
    "editsy": {
      "command": "npx",
      "args": ["-y", "@editsy/mcp"]
    }
  }
}

For agents reading these docs

This site publishes /llms.txt, a plain-markdown map of the docs for tools that read pages as text. The repo's README and architecture notes go deeper than these pages.