Private Journal MCP

Why it's ranked here

This is a focused memory system with unusually practical retrieval. Plain Markdown keeps the source material inspectable, while semantic search makes accumulated notes useful beyond exact keyword matches. Project and global storage scopes fit real engineering work. The main reservation is maturity at the edges: older design documents describe a much smaller tool, and runtime diagnostics expose more environment detail than this privacy-sensitive category warrants.

What's good

Writes remain useful without the server because entries use readable Markdown and YAML metadata. Project notes stay near their codebase, while broader knowledge follows the user across projects. Search runs with local embeddings and supports scope, section, date, and result-count controls. Full-entry reads reject non-Markdown targets, enforce journal roots, resolve symbolic links, and validate the resolved location again before reading.

Tradeoffs

Semantic retrieval adds a transformer dependency, model-loading latency, and separate embedding files beside every entry. Model initialization has a thirty-second timeout, and failed embedding generation leaves the journal entry written but not searchable. Startup diagnostics print home, profile, temporary-directory, user, and journal-location environment values to standard error. Storage relies on normal filesystem permissions, not application-level encryption or access control. Some design documents still describe the earlier write-only interface.

How to use it well

Use it for an agent that repeatedly works with the same person or codebases and should recall decisions, failed approaches, preferences, and technical lessons. Search before complex work, then record concise observations and outcomes afterward. Keep sensitive material within a trusted local account. It does not replace encrypted secrets storage, backups, synchronization, collaborative documentation, or a general project database.

Observed

License

MIT

Primary language

TypeScript targeting Node.js

Packaging and install

npm package with a command-line binary; documented direct execution from GitHub through npx

Interface

stdio-based MCP server with journal writing, semantic search, recent-entry listing, and entry reading tools

Storage

Timestamped Markdown entries with YAML frontmatter and JSON embedding sidecars in project-local or user-global directories

Tests

Repository contains Jest tests for journal writing and path resolution, with the transformer dependency mocked