obra/private-journal-mcp

What it solves

This project provides a private, local-first journaling system that allows AI assistants (specifically Claude) to maintain persistent memory across different conversations. It solves the problem of AI forgetting technical insights, user preferences, and project-specific lessons learned by giving the model a structured way to record and retrieve information.

How it works

The system operates as a Model Context Protocol (MCP) server. It creates markdown files with YAML metadata in either a project-specific directory or a global user directory. To enable discovery, it uses local AI embeddings via @xenova/transformers to perform semantic search, allowing the AI to find conceptually related entries rather than relying on simple keyword matches. All processing and storage happen locally on the machine.

Who it’s for

Developers and users of Claude who want their AI assistant to build a long-term memory of their technical workflows, architectural decisions, and personal reflections without sending sensitive data to external APIs.

Highlights

• Local Semantic Search: Uses local embeddings for natural language queries without external API calls.
• Multi-section Journaling: Organizes entries into categories like reflections, technical insights, and project notes.
• Dual Storage: Separates project-specific notes from global user context.
• Privacy-First: All data and processing remain on the local machine.
• Persistent Memory: Enables AI to track patterns and lessons learned across multiple sessions.