On this page

MCP server

The Model Context Protocol is the integration surface. @skill-wiki/mcp-server-core is a generic, domain-blind MCP server that wraps the runtime and exposes 5 tools.

The 5 tools

Tool	Input	Output	Purpose
`prime_query`	`{ scope, level, tags, budget_tokens }`	`{ hits: [{ id, kind, path, tokens }] }`	Search ranked atoms.
`prime_resolve`	`{ brief, max_tokens }`	+ contracted set	Brief → ranked atoms (L1+L2+L3).
`prime_intent`	`{ brief }`	`IntentObject`	Brief → IntentObject (L1).
`prime_validate`	`{ artifact, atom_set }`	`{ ok, fixes }`	Validate output (L5).
`prime_compile`	`{ source }`	Compiled artifacts	Compile a `.prime` on the fly.

 Boot
 # Standalone (stdio transport)
PRIME_DIR=/abs/path/to/compiled bunx @skill-wiki/mcp-server-core

# Via the prime CLI
prime mcp serve --port stdio --dir /abs/path/to/compiled
 Wire into Claude Code
 {
  "mcpServers": {
    "skill-wiki": {
      "command": "bunx",
      "args": ["@skill-wiki/mcp-server-core"],
      "env": { "PRIME_DIR": "/abs/path/to/compiled" }
    }
  }
}
 See Wire into Claude for the full setup.
 Server architecture
 
   Tool input/output JSON Schema
 Full schemas: Reference → MCP.
 Domain-specific MCP wrappers
  @skill-wiki/mcp-server-core is domain-blind. A wrapper can add
    domain-specific tool names (frontend_color_check), pre-baked tag sets,
    and domain-specific L5 validators. Reference: @skill-wiki/mcp-server-frontend
— see Extending → Domain MCP.