Logseq mcp as safe memory for agents

Back in early 2025 I built Logseq MCP server, recently I rewrote it completely and the concept changed, so sharing it here.

The idea now: Logseq as memory and workspace for AI agents. My graph is not maintained only by me anymore, a couple of Claude Code agents live in it:

  • a nightly agent picks links from my reading list, reads the article (or clones the repo), and writes a structured summary onto the page
  • a morning agent builds a short daily brief from my week plan and tasks
  • every write an agent makes appears as an audit line in my daily journal, so I review their work during my normal morning routine

Safety was the main design goal: agents read broadly (with a page blacklist), but write only inside their own namespace (byAgent/), with a read-before-write gate for edits.

File/Markdown version. GitHub, README has screenshots from my real graph. Feedback welcome!