fitz-graveyard 0.1.0

MCP server for local-first AI architectural planning using local LLMs

fitz-graveyard

MCP server for overnight AI architectural planning using local LLMs.

Features

• Local-first planning with Ollama (Qwen Coder Next 80B/32B)
• Five-stage pipeline: context, architecture, design, roadmap, risk
• SQLite job queue with crash recovery and checkpoint resume
• KRAG-powered codebase context via fitz-ai integration
• Per-section confidence scoring with optional API review
• Cross-platform: Windows, macOS, Linux

Prerequisites

• Python 3.10+
• Ollama installed and running (https://ollama.com)
• Qwen model pulled:

  ollama pull qwen2.5-coder:32b
  

Installation

pip install fitz-graveyard

For API review feature:

pip install "fitz-graveyard[api-review]"

Usage - MCP Server (Claude Code)

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "fitz-planner": {
      "command": "fitz-graveyard"
    }
  }
}

Restart Claude Desktop, then use planning tools directly in conversations.

Available Tools

Tool	Description
create_plan	Queue a new planning job
check_status	Check job progress
get_plan	Retrieve completed plan
list_plans	List all planning jobs
retry_job	Retry a failed job
confirm_review	Approve API review after seeing cost
cancel_review	Skip API review, finalize plan

Configuration

Default configuration location:

• Windows: %APPDATA%\fitz-graveyard\config.yaml
• macOS: ~/Library/Application Support/fitz-graveyard/config.yaml
• Linux: ~/.config/fitz-graveyard/config.yaml

Key settings (YAML):

llm:
  provider: ollama
  model: qwen2.5-coder:32b
  base_url: http://localhost:11434

planning:
  confidence_threshold: 0.75
  max_retries: 3
  checkpoint_interval: 60

api_review:
  provider: anthropic  # Optional: for confidence scoring
  model: claude-opus-4
  enabled: false

Development

git clone https://github.com/yafitzdev/fitz-graveyard.git
cd fitz-graveyard
pip install -e ".[dev]"
pytest

Architecture

fitz_graveyard/
├── server.py           # MCP server entry point
├── engine/             # Planning orchestration
│   ├── stages/         # Five-stage pipeline
│   └── checkpoints/    # Crash recovery
├── queue/              # SQLite job queue
├── context/            # fitz-ai KRAG integration
├── review/             # Optional API review
└── config/             # Configuration management

License

MIT