Praetorian

About

Praetorian is a Model Context Protocol (MCP) server specializing in aggressive context compaction for Claude Code. It dramatically reduces token usage, often by over 90%, by transforming extensive web research, task outputs, and ongoing conversations into structured, incremental snapshots using the efficient TOON (Token-Oriented Object Notation) format. This approach enables frequent and intentional context recycling, allowing developers to maintain fresh, relevant context within their AI coding sessions and rapidly resume work without re-research, thereby significantly enhancing productivity and efficiency.

Key Features

• Includes `praetorian_restore` for searching and injecting relevant context back into the current session.
• Aggressive context compaction providing 90%+ token savings for Claude Code.
• Generates structured incremental snapshots using the efficient TOON format.
• Features `praetorian_compact` for saving and intelligently merging context based on similarity.
• Utilizes Jaccard similarity for smart deduplication and an inverted index for fast keyword search without vector embeddings.
• 3 GitHub stars

Use Cases

• Optimizing context after web research or subagent task completion to reduce token consumption.
• Preserving key decisions and conversational turns for future reference and continuity in projects.
• Maintaining concise and relevant context during long coding sessions to prevent token overload and facilitate instant resumption.