🐕 dotdog

Feed the dog. Ship with specs. Write .dog specs. Dog checks them. AI agents fetch them.

🦴 Getting Started

# npm
npm install -g dotdog

# or Homebrew
brew tap specdog/dotdog
brew install dotdog

dotdog init my-project --minimal    # just SPEC.dog + data-model.dog
dotdog validate                     # score your spec
dotdog compile                      # build .dag graph
dotdog tokens                       # see real byte savings
dotdog serve                        # expose to AI agents via MCP

Adding dotdog to an existing project?

What it does

Describe your app in plain English. dotdog finds what you forgot and fills in the blanks.

$ dotdog init my-app
$ dotdog validate

  my-app : 7 .dog files, 100% complete

$ dotdog analyze

  my-app : 7 files | 100% complete
    SPEC.dog : 5 sections, 1.0KB
    data-model.dog : 3 sections, 1.6KB (5 entities, 6 rels)
  No gaps found.

$ dotdog compile
  ✓ my-app.dag : 3 nodes, 3 edges
  12,400 → 1,860 tokens (85% savings) → real savings: 88.9% smaller

$ dotdog serve
  MCP server running : AI agents query your specs with zero hallucination

Commands

Command	Description
`dotdog validate`	Score spec completeness (0-100%)
`dotdog analyze`	Deep analysis: gaps, entity quality, suggestions
`dotdog parse`	Parse a .dog file into sections
`dotdog compile`	Compile .dog to .dag graph (JSON)
`dotdog generate`	Generate missing spec files from SPEC.dog
`dotdog serve`	MCP server for AI agents over stdio
`dotdog staleness`	Detect drift between spec and reality
`dotdog visualize`	Output Mermaid graph from .dag
`dotdog simulate`	Run a simulation scenario
`dotdog tokens`	Real byte savings (not estimates)
`dotdog init`	Scaffold a new project (`--minimal` for 2 files)
`dotdog list`	List all projects

Format

.dog : Human-written spec (markdown + YAML entities). Free forever.
.dag : Machine-compiled graph (JSON). Token-efficient for AI agents.

MCP Server

dotdog serve runs an MCP server over stdio. AI agents query compiled .dag spec graphs without hallucination.

Setup

Claude Desktop, Cursor, or any MCP client:

{
  "mcpServers": {
    "dotdog": {
      "command": "npx",
      "args": ["-y", "dotdog", "serve"]
    }
  }
}

Run dotdog compile first to generate .dag files.

Tools

Tool	Description
`getEntity`	Full entity: properties, states, lifecycle, connected edges
`traverse`	BFS subgraph from a starting node
`search`	Find entities by name or type
`schema`	Property schema only: names, types, required
`summary`	Node count, edge count, version, token savings
`listProjects`	All project names with compiled .dag files

Example

Agent: listProjects()
→ ["spec-platform"]

Agent: getEntity("spec-platform", "Compile")  
→ { id: "Compile", type: "entity", lifecycle: ["running → completed"], edges: […] }

Agent: summary("spec-platform")
→ { project: "spec-platform", nodes: 11, edges: 5, savings: 91% }

Links

dotdog@0.3.3 · MIT