Skip to main content
Glama

⚔ Houmi MCP & Skills Suite

Autonomous Software Engineering Kernel & High-Craft Design System for Antigravity & Claude Code.

Houmi MCP synthesizes four frontier AI-agent engineering disciplines into a unified, zero-bloat MCP server and slash-command suite:

  1. šŸ‘‘ Fable5 Work Discipline (fable5-mode): Plan-gate (SPEC.md + LEDGER.md), small-card execution, machine-checkable acceptance criteria, and mandatory evidence-on-close.

  2. šŸ›”ļø Godkiller Hard Proof (godkiller): Gates on disk beat chat, zero tolerance for hollow code (stubs/TODOs/mocks), and phase write-guard allowlists.

  3. 🌱 Graft Code Graph Intelligence (Graft): Fast symbol callers, impact radius analysis, and file API skeletons in < 200 tokens.

  4. šŸŽØ Editorial Diagram Design (diagram-design): Generates publication-ready standalone SVG architecture diagrams (Dark Obsidian theme) instead of generic Mermaid boxes.


šŸŽ›ļø Architecture & Core MCP Tools

The houmi-core MCP server exposes 7 focused tools:

ā”Œā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”
│                          houmi-core MCP (7 tools)                        │
ā”œā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”¤
│  🧭 houmi_route          Auto-detect intent → mode + subsystem triggers  │
│  šŸ“‹ houmi_ledger         Fable5 task cards on disk (.agents/LEDGER.md)   │
│  šŸ”— houmi_backend_sync   Backend API / Tauri IPC contract manager        │
│  šŸ”„ houmi_stack_grill    React 19 / Fabric.js / TypeScript / Token audit │
│  🌳 houmi_graft_intel    Symbol callers (blast radius) | file skeleton   │
│  šŸŽØ houmi_diagram        Editorial SVG diagrams (Dark Obsidian style)    │
│  šŸ›”ļø houmi_guard          Write-guard path allowlisting per phase         │
ā””ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”˜

Related MCP server: SnapBack

šŸš€ Slash Commands

Command

Mode

Purpose

/houmi

🧠 Master Auto-Detect

Single entry point: auto-routes Canvas, Backend sync, Diagrams, and Grilling

/loop

šŸ”„ Marathon Loop

Continuous 6–8 hour autonomous cycle with self-grilling and auto-wake

/plan

šŸ“ Plan & Spec

Phased plan, Editorial SVG diagram, Fable5 LEDGER cards

/ask

šŸ” Read-Only Explore

Graft symbol search and API skeleton (< 200 tokens)

/debug

šŸ› Diagnosis

Hypothesis → blast radius → minimal fix → stack verify

/ultradeep

⚔ 1-Phase Execution

Single card execution with write-guard

/view

šŸ‘ļø Deep Architecture

Architecture study and editorial system diagrams

/verify

āœ… Evidence Gate

Machine-checkable build test and LEDGER close

/jury

šŸ›ļø Final Audit

Pre-release compliance check


šŸ“¦ Quickstart & Installation

Option 1 — One-Line PowerShell (Windows)

powershell -ExecutionPolicy Bypass -File .\scripts\install.ps1

Option 2 — One-Line Bash (Linux / macOS)

bash ./scripts/install.sh

Option 3 — Manual Setup

# 1. Install package in editable mode
pip install -e .

# 2. Add to your mcp_config.json (e.g. ~/.gemini/antigravity/mcp_config.json):
{
  "mcpServers": {
    "houmi-core": {
      "command": "houmi-mcp"
    }
  }
}

# 3. Copy skills to your global or project skills directory:
cp -r skills/* ~/.agents/skills/

šŸ“„ License

MIT Ā© NinjaScoutZ

Related MCP Connectors

Related MCP Servers

  • A
    license
    Not graded
    quality
    C
    maintenance
    Enables local-first task contract generation and drift/sludge guard checks for coding tasks, with CLI and VS Code extension support.
    38 npm
    MIT
  • A
    license
    Not graded
    quality
    A
    maintenance
    Enables agentic coding tools to maintain a living, in-repo model of intent, architecture, and code, with deterministic indexing, bidirectional linking, and continuous drift detection for task-scoped context and reconciliation.
    MIT