Commands — Engrams Docs

Setup

Add Engrams to your Claude project once. Every session after that boots automatically with full context.

Setup Add to Claude Project Instructions

▼

Paste this into your Claude.ai Project Instructions. Replace YOUR_API_KEY with the key from your Engrams dashboard. Claude will read your CLAUDE.md on every boot automatically.

Claude.ai — Project Instructions

Fetch and follow: https://raw.githubusercontent.com/YOUR_ORG/YOUR_REPO/main/CLAUDE.md

Engrams API key: YOUR_API_KEY
Engrams endpoint: https://engrams.app/api/mcp

⚠️ Keep your API key private. Never commit it to a public repo.

Setup CLAUDE.md — minimal template

▼

Create a CLAUDE.md in your repo root. This is the file Claude reads at every session boot. Customize it for your project.

CLAUDE.md — minimal

# CLAUDE.md — [Your Project]

## Session Boot Protocol
1. Read `state/session_handoff.md`
2. Read `state/work_queue.md`
3. Summarize status and propose next steps

## Session Handoff Protocol
1. Update `state/session_handoff.md`
2. Update `state/work_queue.md`
3. Log decisions via Engrams `log_decision`
4. Write session via Engrams `write_session`
5. Commit and push: `git add state/ && git commit -m "state: handoff" && git push`

Session commands

These three commands form the core loop. Boot loads context, handoff saves it, sync keeps parallel sessions aligned.

Boot session boot

▼

Loads full project context at the start of every session. Reads your CLAUDE.md, last session handoff, work queue, and any agent briefings. Claude presents a status summary and proposed next steps before you say anything.

When to use

Automatically on every new session if your CLAUDE.md includes the boot protocol. Also useful to re-run mid-session after a long break.

Say to Claude

session boot

What Claude reads

state/session_handoff.md     — what happened last session
state/work_queue.md          — prioritized task list
state/daily_briefing.md      — agent briefing (if exists)
project_memory/goals.md      — project objectives
CLAUDE.md                    — your project instructions

Tip: In Claude.ai, paste the raw URL to your CLAUDE.md in Project Instructions. Claude fetches it fresh every session — no UI updates needed.

Handoff session handoff

▼

Saves everything that happened this session so the next one can pick up exactly where you left off. Updates handoff file, work queue, logs decisions, writes to Engrams memory, and commits to GitHub.

When to use

Before ending any session. Before switching tools (e.g. Claude.ai → Claude Code). When Claude warns the context window is getting full.

Say to Claude

session handoff

What Claude writes

state/session_handoff.md     — summary, decisions, next steps
state/work_queue.md          — updated task list
project_memory/decisions.md  — structural decisions log
state/active_context.md      — live context for other tools

Then commits

git add state/ project_memory/ CLAUDE.md && git commit -m "state: session handoff YYYY-MM-DD" && git push

Claude.ai and Claude Code both run handoff. GitHub is the single source of truth that connects them.

Sync sync

▼

Fetches the latest context from Engrams into an active session. Use this when Claude.ai has updated something and you want Claude Code (or another tool) to know about it — without ending the session.

When to use

In Claude Code when a parallel Claude.ai session has made decisions. At the start of a Claude Code session to check for updates. Any time you’re unsure if your session has the latest context.

Install alias (add to ~/.zshrc)

alias sync='curl -s https://raw.githubusercontent.com/YOUR_ORG/YOUR_REPO/main/state/active_context.md'

In Claude Code — say

sync

Or run directly

curl -s https://raw.githubusercontent.com/YOUR_ORG/YOUR_REPO/main/state/active_context.md

Claude.ai writes to state/active_context.md during handoff. This file is the live whiteboard between all your AI tools.

Memory commands

Direct API calls to the Engrams memory layer. Claude calls these automatically as part of handoff — but you can also trigger them explicitly.

Memory log_decision

▼

Logs a structural decision to Engrams memory. Every decision gets a timestamp, rationale, and impact. Claude can surface these in future sessions so it never forgets why something was built a certain way.

Say to Claude

log_decision: [what was decided] because [why] — impacts [what]

API call (Claude handles this)

POST https://engrams.app/api/mcp
{
  "tool": "log_decision",
  "input": {
    "decision": "Switched auth from JWT to sessions",
    "rationale": "JWT expiry caused UX issues on mobile",
    "impact": "All API endpoints now require session cookie"
  }
}

Memory write_session

▼

Writes a session summary to Engrams memory. Called automatically during session handoff. Creates a searchable log of what was built, decided, and left open across all sessions.

API call

POST https://engrams.app/api/mcp
{
  "tool": "write_session",
  "input": {
    "summary": "Built auth flow, fixed session bug",
    "changes": ["Added /api/login", "Fixed cookie expiry"],
    "next_steps": ["Build /api/logout", "Add rate limiting"],
    "project_phase": "build",
    "energy": "momentum",
    "agent_id": "claude-ai"
  }
}

Memory read_session

▼

Retrieves the last N sessions from Engrams memory. Used during boot to give Claude full historical context, not just the last session.

Say to Claude

read last 5 sessions

API call

POST https://engrams.app/api/mcp
{
  "tool": "read_session",
  "input": { "limit": 5 }
}

Memory restore_session

▼

Restores context from a specific past session. Useful when you want to pick up from a point further back than the last handoff — for example, before a big refactor.

Say to Claude

restore session from 2026-03-25

API call

POST https://engrams.app/api/mcp
{
  "tool": "restore_session",
  "input": { "date": "2026-03-25" }
}

Agent commands

Commands for autonomous agents that run on a schedule and write to your memory layer.

Agent next_boot_preview

▼

Writes a pre-computed briefing that Claude reads at the next session boot. Agents use this to surface market data, task updates, or status reports before you even say a word.

API call (run by agent)

POST https://engrams.app/api/mcp
{
  "tool": "next_boot_preview",
  "input": {
    "briefing": "Market: RISK-OFF. 3 new top gainers. AGX +37%.",
    "priority_items": ["Review AGX position", "Check SECTOR_HOT agents"]
  }
}

Agent health

▼

Returns the health status of your Engrams connection — API reachable, memory readable, last write timestamp. Run this to verify your setup is working.

Say to Claude

engrams health check

Direct call

curl https://engrams.app/api/mcp/health \
  -H "Authorization: Bearer YOUR_API_KEY"

Response

{
  "status": "ok",
  "project": "my-project",
  "last_write": "2026-03-28T10:14:00Z",
  "sessions_stored": 24
}