dotdog

CLI and MCP tooling for structured .dog specs, compiled .dag graphs, repo mapping, and multi-repo workspaces.

AI agents are everywhere now — in your editor, your terminal, your CI pipeline. They know every framework and every API. But they do not know your project. They hallucinate entity names, invent relationships, and miss required fields because they are reading prose and guessing.

dotdog fixes this. You write your spec in readable .dog files — entities, properties, states, lifecycles, relationships. dotdog compiles them into a .dag graph that is 94% smaller than the source. The agent loads the graph instead of scanning 600 lines of prose. It queries exact data instead of interpreting English. Hallucination drops to near zero.

Install

npm install -g dotdog    # npm
brew install dotdog       # Homebrew
bun add -g dotdog         # bun

Requires Node.js >= 20 or Bun >= 1.3.

See it work

$ dotdog init my-project
  Created 5 .dog files in projects/my-project/

$ dotdog validate
  my-project : 5 .dog files, 95% complete

$ dotdog compile
  ✓ my-project.dag
    5 nodes, 3 edges
    4620 → 280 tokens (93.9% savings)

$ dotdog serve
  MCP server ready — 6 tools available

The DAG is 94% smaller than source. Agents load the full entity graph in one shot instead of scanning prose. The DAG does not replace reading; it replaces scanning.

How it works

You write specs before code. Five minutes to set up. Zero configuration.

Step	Command	What happens
Scaffold	`dotdog init`	Creates SPEC.dog, data-model.dog, and supporting files
Describe	edit `.dog` files	Define entities, properties, states, lifecycles, relationships
Validate	`dotdog validate`	Score completeness. Find missing entities and broken links
Compile	`dotdog compile`	Build a positional DAG graph — 94% smaller, optimized for LLM context
Design	`dotdog design`	Find missing data-model decisions and concrete next steps
Expose	`dotdog serve`	Start a local stdio MCP server. AI agents query via nine structured tools

Observed workspace graph

Dotdog can also observe an existing repo or multi-repo workspace and write deterministic graph artifacts:

dotdog observe
dotdog ask "which files define routes?"
dotdog drift

observe writes .doghouse/observed.json, .doghouse/facts.jsonl, and .doghouse/workspace.dag. ask queries those facts without an LLM dependency. drift reports stale or missing observed references.

Generated .doghouse graphs and observed facts are ignored by Git by default because they may contain repository metadata. Workspace output uses repository-relative paths.

For AI agents

Nine MCP tools for structured queries — no scanning, no guessing:

getEntity · traverse · search · path · schema · summary · listProjects · workspace.list · infraVerify

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

The bundled server uses stdio, opens no TCP listener, and writes no query logs. Secure any external gateway separately.

AI agents: read the for agents · dogfood page before working with dotdog projects.

Learn more

Spec-Driven Development guide — the methodology behind the tool
Tutorial: Build a spec-driven project
Contributing Kits
FAQ

dotdog@0.8.1 · MIT · GitHub · npm