Skip to content

quickstart.md

Latest commit

 

History

History
156 lines (104 loc) · 4.02 KB

quickstart.md

File metadata and controls

156 lines (104 loc) · 4.02 KB
Raw

Quickstart

Get Lorekeeper running and seeding memories in ~2 min (warm cache) or ~5 min on first install (sentence-transformers model download is ~90 MB).

Prerequisites

  • Python 3.11+ (uv python install 3.11 if missing)
  • uv — curl -LsSf https://astral.sh/uv/install.sh | sh
  • One of: Hermes Agent, Claude Code, or Cursor

1 — Install & configure

pip install lorekeeper-mcp
lorekeeper setup

If you are working from a git checkout, bash scripts/setup.sh still exists and also installs

development hooks/skills. For the primary PyPI install path, lorekeeper setup is the command to run.

What this does:

  • Installs the package and bundled CLI
  • Creates ~/.lorekeeper/ (data directory)
  • Auto-detects Hermes / Claude Code / Cursor and injects MCP config + prompt section
  • Installs skills for each detected agent
  • Prints a seed prompt — keep the terminal open for step 5

2 — Restart your agent

After lorekeeper setup finishes, restart your agent (Claude Code, Hermes, or Cursor) to load the new MCP config. The lorekeeper MCP server starts automatically on the next agent launch.

3 — Start the dashboard

In a separate terminal:

uv run lorekeeper-dashboard

Then open: http://127.0.0.1:7777

The dashboard starts empty — memories appear here in real time as you insert them.

Change the port: LORE_DASH_PORT=8888 uv run lorekeeper-dashboard

4 — Seed your first memories

Copy the seed prompt printed at the end of lorekeeper setup and paste it into your agent:

Read your prompt/config files (soul.md, CLAUDE.md, .cursorrules, AGENTS.md) and
save key facts about yourself to Lorekeeper using lore_remember or lore_insert — who you
are, what you do, your constraints and preferences. Be thorough.

The agent will call lore_remember / lore_insert and the dashboard will populate.

5 — Verify MCP round-trip

In your agent, run these two back-to-back:

lore_remember("My first test memory")
lore_search("test memory")

Expected lore_search result — you should see the memory you just stored:

{
  "memories": [
    {
      "title": "My first test memory",
      "combined_score": 0.87
    }
  ],
  "count": 1
}

CLI flags

The lorekeeper binary now accepts --help and --version without starting the MCP server:

lorekeeper --help
# usage: lorekeeper [-h] [--version] {setup} ...
# Personal AI memory MCP server — stores facts and knowledge for AI agents.

lorekeeper --version
# lorekeeper 0.2.0

uv run lorekeeper --help works too if the binary isn't on PATH.

Troubleshooting

lorekeeper: command not found

The binary is only on PATH if installed via uv tool install or a virtualenv. Use uv run lorekeeper from inside the repo, or run:

uv tool install ./dist/lorekeeper_mcp-0.2.0-py3-none-any.whl

Dashboard is empty after seeding

  1. Check that the agent actually called lore_remember / lore_insert (look at the agent's tool call log).
  2. Confirm the MCP server connected — if not, re-run lorekeeper setup and restart the agent.
  3. Hard-refresh the dashboard (Cmd+Shift+R).

MCP tools not available in agent

  • Confirm lorekeeper setup ran without errors and showed your agent in the summary table.
  • Check the agent's MCP config file for a lorekeeper entry:
    • Claude Code: ~/.claude/settings.json
    • Cursor: ~/.cursor/mcp.json
    • Hermes: ~/.hermes/mcp.yml
  • Restart the agent if you edited config manually.

Next steps

  • Browse all available tools: [lore_search, lore_remember, lore_insert, lore_update, lore_forget, lore_reflect, lore_recommend_links, lore_processed_sessions]
  • Full configuration reference: README.md — Configuration
  • Architecture deep-dive: docs/ARCHITECTURE.md