AI Build Journal
An MCP server that lets Claude read and write to a structured daily journal of what I built. Designed for the agent-first workflow — the journal is the agent's memory.
An MCP (Model Context Protocol) server that gives Claude a structured, searchable journal of what I built, what I learned, and what I got stuck on. The journal is stored in SQLite, queryable through MCP tools, and exposed as a static daily log I can browse.
Why
When I work with Claude Code, the agent has no memory between sessions. I end up re-explaining context every time I start a new task. This server is the agent’s long-term memory: when I ask Claude “what was I working on last week?” or “have I seen this error before?”, it queries the journal instead of guessing.
What it does
- Exposes MCP tools:
journal_today,journal_search,journal_log,journal_recent. journal_logaccepts a free-form note from the agent and stores it with timestamp and tags.journal_searchis a simple LIKE-based search; full-text search is a v0.4 milestone.- Backed by a single SQLite file, rotated weekly.
Stack
Python · FastMCP · SQLite · systemd. Runs on the Hetzner VPS, reachable only via Tailscale. The journal file is the canonical record; the MCP server is a thin wrapper.
Auth
Tailscale-only. The journal is personal; no one else needs to read it, and exposing it publicly would be a privacy regression. The MCP server is reachable only on my tailnet.
Stack
python · fastmcp · sqlite · systemd
Auth
Tailscale-only. The app is reachable only on the author's tailnet.