dci-mcp

An MCP server that generates code adhering to the DCI architecture when generating or refactoring code. It loads language-specific rules and examples, then instructs the LLM to apply them immediately — no back-and-forth.

Tools

| Tool | When to use | | -------------------------------- | -------------------------------------------------------------------------- | | prepare_dci_refactor | User wants to refactor existing code into DCI | | scaffold_dci_from_mental_model | User describes a mental model / user story and wants DCI code from scratch |

Both tools accept a language argument (e.g. "typescript", "javascript") and return the full DCI ruleset for that language as context.

Supported languages

Add a folder under docs/ with instructions.md (required) and examples.md (optional):

docs/
  core.md              # DCI rules shared across all languages
  typescript/
    instructions.md
    examples.md
  javascript/
    instructions.md
    examples.md

Usage in mcp.json (stdio)

"dci": {
  "command": "pnpx",
  "args": ["dci-mcp"]
}

Or with npx:

"dci": {
  "command": "npx",
  "args": ["-y", "dci-mcp"]
}

Development

pnpm install
pnpm dev          # build + pnpm link (makes dci-mcp available globally)
pnpm inspector    # open MCP Inspector connected to the local server

After pnpm dev, the server is available as dci-mcp in your PATH and can be used in mcp.json outside this project while you iterate:

"dci": {
  "command": "dci-mcp"
}