MemoryVault — Team Onboarding

Path A — Ask via the hosted connector (easiest)

Zero local setup — query the vault from Claude.ai (browser)

Add one connector to your Claude.ai profile and ask questions against the org vault. It's always the latest published version (re-indexed within ~15 min of any update), and there's nothing to install or maintain. Tools exposed: get_memory / search_vault / get_page / list_pages.

1. Open https://claude.ai/settings/connectors
2. Click Add custom MCP
3. URL: https://memoryvault-mcp.tradeit.gg/mcp  ·  leave OAuth Client ID/Secret blank (dynamic registration)
4. Click Connect → complete Google SSO with your @zengaming.com / @tradeit.gg account

Use the memoryvault tool to search for "pricing domain" and summarize the top result.

 "That account does not have access" → your Google domain isn't allowlisted; ping the maintainer. Already fixed but still denied? Cloudflare Access caches the session — open https://tradeitgg.cloudflareaccess.com/cdn-cgi/access/logout and retry.

 Want it inside Claude Code, faster, and working without a per-query round-trip? → Local team vault (next tab).

Path B — Local team vault (developers)

A local copy + a hook that feeds Claude Code every session

Clone the vault to your machine. A SessionStart hook injects MEMORY.md into every Claude Code session — so the org context is already in-context with no per-query network call (faster than the connector), and it auto-pulls the latest daily.

Prereq What you need

brew install git jq gh && brew install --cask claude-code
gh auth status   # must show you're logged in to zengamingx; else: gh auth login

winget install --id Git.Git -e          # Git for Windows = git + Git Bash
winget install --id GitHub.cli -e       # gh
winget install --id jqlang.jq -e        # jq
winget install --id Anthropic.ClaudeCode -e
# open a NEW terminal so PATH refreshes, then:
gh auth status   # must show you're logged in to zengamingx; else: gh auth login

Step 1 Install

gh repo clone zengamingx/memoryvault-org ~/MemoryVault-team \
  && bash ~/MemoryVault-team/install.sh

gh repo clone zengamingx/memoryvault-org ~/MemoryVault-team

bash ~/MemoryVault-team/install.sh --only-hook

mkdir "%USERPROFILE%\.memoryvault" 2>nul
schtasks /create /tn "MemoryVault Team Pull" /sc daily /st 09:30 /f ^
  /tr "cmd /c git -C \"%USERPROFILE%\MemoryVault-team\" pull --ff-only origin main >> \"%USERPROFILE%\.memoryvault\pull.log\" 2>&1"

Step 2 Teach your Claude — reader contract (read-only)

Append to your ~/.claude/CLAUDE.md (create the file if needed):

 Before pasting, replace every <your-repos-dir> below with the directory where you clone tradeit repos — e.g. ~/Projects, ~/work, or ~/dev.

# Team MemoryVault — Reader Contract (read-only)

You have a read-only, auto-refreshed copy of tradeit.gg org knowledge at
`$HOME/MemoryVault-team`. The SessionStart hook injects `MEMORY.md` every session.
It is the source of truth for org-level synthesis — repo overviews, infra
patterns, MySQL/Redis hard rules, deploy conventions, domain vocabulary.

## This vault is READ-ONLY
`~/MemoryVault-team` is overwritten by a daily `git pull`. NEVER write, edit, or
commit files there — changes are lost on the next pull. If you learn something
durable about tradeit.gg, tell the user to relay it to the vault maintainer so
it gets published to everyone.

## Read order
1. `MEMORY.md` — already in context; orientation + high-signal entry pages.
2. `index.md` — full catalog, when MEMORY.md doesn't cover the topic.
3. Specific `wiki/**/*.md` pages — via wikilinks `[[slug]]`.

## Frontmatter is authoritative
`confidence < 0.6` → speculative, verify. `last_verified` > 90 days → stale, say
so. `superseded_by: ` → read the successor, not this page.

## Repo-fact routing (LOAD-BEARING)
_Replace `<your-repos-dir>` with wherever you clone tradeit repos (e.g. `~/Projects`)._
- How to work IN repo X (build/test/in-repo conventions) → `<your-repos-dir>/X/CLAUDE.md` ONLY.
- What X IS / how it fits / cross-repo synthesis → `wiki/entities/repo-X.md`.
- Cross-repo conventions (MySQL, git, infra) → `wiki/concepts/*.md`.
- On conflict, the per-repo `CLAUDE.md` wins (closer to the code); flag the vault page as stale.

## Never echo secrets
If you ever see an inline credential/token in a page, flag it — don't repeat it.

Step 3 Verify

ls ~/MemoryVault-team/wiki/entities | head
launchctl list | grep memoryvault
tail ~/Library/Logs/memoryvault-team-pull.log

dir "%USERPROFILE%\MemoryVault-team\wiki\entities"
schtasks /query /tn "MemoryVault Team Pull"
type "%USERPROFILE%\.memoryvault\pull.log"

Day-to-day

 You can also add the browser connector (Path A) for Claude.ai access — the two coexist.

Path C — Personal vault (optional)

1. Bootstrap your personal vault

export VAULT_DIR="$HOME/MemoryVault"
mkdir -p "$VAULT_DIR"/{wiki/entities,wiki/concepts,wiki/sources,wiki/topics,raw,scripts,archive/inbox,.inbox,.export,assets}
cd "$VAULT_DIR" && git init -q && echo ".export/" > .gitignore

# Copy the hook + lint scripts + schema spec from the skeleton (ask Ehud):
SRC="/path/to/memoryvault-skeleton"
cp "$SRC/scripts/"{session-start-hook.sh,session-stop-hook.sh,export-org.sh,lint.sh,lint-repo-entities.sh} "$VAULT_DIR/scripts/"
cp "$SRC/CLAUDE.md" "$VAULT_DIR/"
chmod +x "$VAULT_DIR/scripts/"*.sh

Prereqs for this path: brew install node and brew install --cask obsidian.winget install --id OpenJS.NodeJS -e and winget install --id Obsidian.Obsidian -e. Run the bootstrap block above in Git Bash (the mkdir -p brace-expansion and $HOME are bash).

2. Wire the personal hooks

Append to ~/.claude/settings.json (alongside the team hook from Path B — both run):

{
  "hooks": {
    "SessionStart": [
      { "hooks": [ { "type": "command", "command": "bash $HOME/MemoryVault/scripts/session-start-hook.sh" } ] }
    ],
    "Stop": [
      { "hooks": [ { "type": "command", "command": "bash $HOME/MemoryVault/scripts/session-stop-hook.sh" } ] }
    ]
  }
}

3. Obsidian + personal reader contract

Open $HOME/MemoryVault as an Obsidian vault (enable [[Wikilinks]] + the Dataview plugin). For a personal vault Claude SHOULD write back, so use a write-back reader contract (it creates/updates pages under $HOME/MemoryVault/wiki/, updates index.md/log.md, surfaces .inbox/, suggests /memoryvault-revise at session end). Ask Ehud for the full personal-vault reader-contract block — it's the same one his master vault uses.

Path D — Use the vault as Claude's memory (advanced)

1. Archive the existing default memory (avoid split-brain)

Move the current per-project memory aside so Claude doesn't read two stores. Nothing is deleted:

for d in ~/.claude/projects/*/memory; do
  [ -d "$d" ] && mv "$d" "$d.archived-$(date +%F)" && echo "archived $d"
done

 Run this in Git Bash — it's a bash loop. Your profile path is the same one Git Bash shows as ~ (i.e. %USERPROFILE%).

2. Add the memory-routing block to ~/.claude/CLAUDE.md

# Memory routing — personal vault supersedes default memory

My durable memory lives in my personal MemoryVault at `$HOME/MemoryVault`, NOT in
Claude Code's default `~/.claude/projects//memory/` files (archived).
(The read-only team vault at `~/MemoryVault-team` is a separate feed — never write there.)

- READ: the personal vault's `MEMORY.md` (injected by its SessionStart hook) is the entrypoint.
- WRITE: durable facts about me/my work → a page under
  `$HOME/MemoryVault/wiki/{entities,concepts,sources,topics}/` with frontmatter
  (scope, confidence, last_verified); update `index.md`, append `log.md`. The Stop
  hook auto-commits it.
- DO NOT write durable facts to `~/.claude/projects//memory/` — retired;
  files there won't load in other projects/sessions.
- Ephemeral, task-only notes can stay in the default project memory.
- Check for an existing page before creating a duplicate. Never save secrets.

Same routing Ehud's global CLAUDE.md uses — ask him for the fuller version (scope/category rules) if you want it.

3. Verify

Fresh session → ask Claude to remember something durable about you. It should propose a page under $HOME/MemoryVault/wiki/… (Stop hook commits it), not a write to ~/.claude/projects/…/memory/.