#!/usr/bin/env python3

-- coding: utf-8 --

""" Plan Version Control (PVC) MCP Server — Quick Start

A Git-like version control system for AI agent decision tracking. Traditional agents remember chat content. PVC agents remember the decision evolution process.

Installation

pip install pvc-mcp-server

Related MCP server: elefante

CLI Usage

# Start the MCP server
pvc-mcp

# Or as a Python module
python -m pvc_mcp

Claude Desktop Configuration

Add to your claude_desktop_config.json:

.. code-block:: json

{
  "mcpServers": {
    "pvc": {
      "command": "python",
      "args": ["-m", "pvc_mcp"]
    }
  }
}

Python API

.. code-block:: python

from pvc_mcp import plan_commit, plan_log, plan_diff

# Commit a plan
result = plan_commit(
    plan={"goal": "Build a blog system", "tasks": [...]},
    message="Initial plan"
)

# View history
history = plan_log(limit=10)

# Compare versions
diff = plan_diff("v1", "v2")

Core Concepts

PVC is like Git for AI agent plans:

============ =========== =========================== PVC Tool Git Analog Purpose ============ =========== =========================== plan_commit git commit Save a plan version plan_diff git diff Compare two versions plan_checkout git checkout Restore a historical version plan_branch git branch Create an experimental branch plan_merge git merge Merge branches (3-way) plan_log git log View decision history plan_tag git tag Mark milestones plan_stash git stash Temporarily shelve changes ============ =========== ===========================

Storage Structure

::

.plans/
├── HEAD                 # Current pointer
├── refs/heads/          # Branch references
├── refs/tags/           # Tag references
├── objects/             # Immutable plan snapshots
└── logs/                # Operation history

License

Apache 2.0 — see LICENSE file.