Intent & Claims
Declare what you plan to change, then lock files while editing. No more surprise merges or conflicting changes.
SCRUM (Synchronised Claims Registry for Unified Multi-agents) is the coordination layer that keeps every agent honest and in sync. Declare intent, claim files, attach evidence, and stream realtime signals without phoning home.
New in v0.5.2
Convenience tools combine common workflows into single calls. What used to take 10+ tool calls now takes 3.
scrum_context()
Session startup in ONE call. Returns status, tasks, claims, board.
scrum_start_work()
Begin work in ONE call. Does overlap → intent → claim.
scrum_finish_work()
Complete work in ONE call. Does evidence → compliance → release.
Also new: Pre-commit hooks · Profile auto-detection (solo/team/full) · Orchestration tools · Sprint collaboration · Unified blocker/dependency/metrics tools · Compliance verification
Start here
Install SCRUM locally, point your agents at the MCP server, and watch intent, claims, and receipts appear in one shared lobby.
npm install
npm run dev
npm run build
npm run mcp
API: http://localhost:4177/api
WebSocket: ws://localhost:4177/ws
Dashboard: http://localhost:5174
Dashboard requires cd frontend && npm install && npm run dev
npm run build
npm run cli -- status
npm run cli -- task create --title "First task"
The coordination surface
SCRUM MCP is not another dashboard. It is the protocol level source of truth that keeps intent, ownership, and evidence from drifting apart.
Declare what you plan to change, then lock files while editing. No more surprise merges or conflicting changes.
Multiple agents working on the same task share context in real-time. Decisions, interfaces, questions—all coordinated.
Verify agents did what they declared. Block releases on undeclared files or boundary violations.
Attach command output as proof. If it is not evidenced, it did not happen. Full audit trail.
Full board with dependencies, WIP limits, blockers. Track cycle time, velocity, and aging work.
SCRUM never phones home. Runs where your repo lives. Never edits without consent. MIT licensed.
MCP built in
SCRUM exposes its coordination layer as MCP tools so Claude Code, Cursor, Continue, and any MCP client can share intent and receipts instantly.
npm install
npm run build
npm run mcp
# MCP config example
{
"mcpServers": {
"scrum": {
"command": "node",
"args": ["/absolute/path/to/scrum/dist/mcp.js"],
"env": { "SCRUM_DB_PATH": "/path/to/.scrum/scrum.sqlite" }
}
}
}
A tighter loop
SCRUM enforces a workflow that prevents silent failures and forces clarity across every agent.
Declare scope and acceptance criteria before edits begin.
Hold short-lived ownership while you work.
Capture command output and receipts for every change.
Unlock claims only after evidence is attached.
Enforce at commit
Pre-commit hooks validate SCRUM compliance before every commit. If you didn't declare intent, didn't claim files, or violated boundaries—the commit is blocked.
# Install the pre-commit hook
./scripts/install-hooks.sh
# Set your task context
export SCRUM_TASK_ID="task-abc123"
export SCRUM_AGENT_ID="claude-code-xyz"
# Commit as normal - hook validates compliance
git add src/feature.ts
git commit -m "Add feature"
# ✅ SCRUM compliance passed (score: 100/100)
# Or if you forgot to declare files...
# ❌ COMPLIANCE FAILED: Undeclared files modified
Ops ready
Operations should not guess what happened at 3AM. SCRUM keeps a live ledger of tasks, intents, claims, and evidence.
Check /api/status and the event stream for drift.
Roll back by restoring the last known good build.
Inspect evidence receipts before you merge.
SCRUM never auto-edits your repo, never phones home, and logs what it is doing and why. Anything destructive is behind an explicit command.
Security postureShip with evidence
SCRUM MCP keeps the lights on for every agent, every task, every deploy. It is the coordination layer your team assumes already exists.