MCP integration

The same reading capabilities behind the c4s commands are also available over MCP through a companion binary, c4s-mcp. It runs a read-only server over stdio, so any MCP-capable agent can browse your spec — list entities, fetch details, follow references — without you wiring up the commands yourself.

The editor already does this for the agent inside it: the first run writes a .claude4spec/mcp.json registering this server. You only reach for c4s-mcp by hand when you want a spec available to an agent outside the editor — your own Claude Code setup, a script, or another tool.

Register the server

Add c4s-mcp to your MCP client's server config. Point --project at the spec you want to expose:

{
  "mcpServers": {
    "c4s-reader": {
      "command": "c4s-mcp",
      "args": ["--project", "/path/to/your/spec"]
    }
  }
}

If you omit --project, c4s-mcp walks up from the working directory to find the nearest .claude4spec/ — the same resolution the c4s commands use.