Two problems: copy-paste ops and decision theater

Project teams face dual pain. Repetitive work—CRM won → create project, ticket tagged → spawn task, signed doc → unlock column—still done by hand between systems. Hard decisions—delay vs descope, hire vs cut scope—redebated every sprint without logged forecasts.

Mature organizations separate machine work from judgment work. WKFGo supports both: n8n and MCP write tools (create_task, update_status, log_activity, create_wiki_page) for repeatable glue; simulate_scenario, executive_brief, decision_inbox, and boardroom patterns for forks humans must own.

Part 1 — Repetitive work automation (n8n + MCP)

What to automate

High-ROI flows share traits: clear trigger, stable mapping, idempotent outcome.

Trigger	Action via n8n or MCP
CRM deal closed	Create project from template; assign PM
Form submission	Triage task + notify channel
Desk ticket escalated	Link task; set priority
Document signed	Move gating task; log activity
Scheduled overdue check	Nudge owner; optional inbox item

n8n excels at HTTP orchestration, schedules, and transforms—self-hosted or cloud. WKFGo REST and MCP endpoints become nodes in visual workflows. Sidebar links (N8N_URL) connect operators from the app shell to their canvas.

MCP writes suit agentic assistants: "Create onboarding tasks for hire X" after HR confirms—human approves batch, bot executes create_task calls with service credentials scoped by feature access.

Design rules

Idempotency — store external IDs on tasks; retries must not duplicate projects
Failure visibility — n8n error branches ping humans; silent failure worse than manual work
Permissions — service accounts minimal scope; no superuser tokens in shared workflows
Document in wiki — create_wiki_page listing automations and owners

n8n vs in-product BPMN

WKFGo BPMN models human sequences—approvals, reviews, role handoffs with audit. n8n syncs systems without clicks. Use BPMN when people must act in order; use n8n when APIs must talk overnight.

Part 2 — Decision scenarios (simulate + boardroom + briefs)

Automation speeds execution; it does not pick release dates. For constraint-changing choices:

simulate_scenario

Compare explicit futures:

delay_release — date moves, scope intact
cut_scope — date holds, features defer
add_people — capacity injection with cost assumptions
freeze_project — portfolio WIP control

Output spans dimensions returned by the tool—state assumptions; do not invent ROI slides.

executive_brief and boardroom

Pull officer lenses (cfo, cto, cpo, …). CEO synthesis exposes board.tensions when officers would disagree. Boardroom prompt pattern: each states FACTS-linked position; CEO resolves or escalates to simulation.

decision_inbox loop

Background monitors raise items when signals cross thresholds—consolidated Executive Digest, not ping storms. Workflow:

decision_inbox — prioritized queue
Evidence from briefs and reports
simulate_scenario on top forks
resolve_decision_item + log_decision forecast
decision_log calibration next cycle

Combined architecture (how both halves fit)

External systems ──n8n──► WKFGo tasks/wiki (automation)
                              │
Portfolio signals ──monitor──► decision_inbox
                              │
Leaders ──brief/simulate/boardroom──► logged decisions
                              │
MCP agents ──read/write──► same source of truth

Agents and n8n share one database truth—automation creates tasks; decisions change which tasks matter.

Example week

Monday — n8n creates projects from CRM wins; MCP standup digest posts movement
Wednesday — monitor raises inbox item on burn threshold; CFO brief pre-read
Thursday — boardroom debate cut_scope vs delay; simulate_scenario both; CEO resolves
Friday — n8n sends stakeholder email draft trigger after human approval—not auto-send

Governance: who may write what

Separate read MCP keys (search, briefs, reports) from write keys (create_task, update_status, log_decision). n8n service accounts typically write operational objects; only facilitators publish wiki or resolve inbox items. Demo users remain read-only—automation must not bypass demo middleware.

Review automation quarterly in the same meeting as decision_log calibration: Did n8n create orphan tasks? Did inbox resolutions match implemented board state? Drift between automated writes and human decisions indicates broken handoff design—not "AI error."

Anti-patterns

Automating executive approvals without audit
Running simulate once, never logging choice
MCP superuser in Slack—leaks finance tasks
n8n creating tasks without assignees
Using automation to bypass demo read-only guards (blocked by design)

Automate provisioning and reminders; never automate executive approval without audit trail. log_activity on MCP writes preserves who changed what for compliance reviews.

Run a monthly audit: do AI answers cite task IDs from MCP? If not, tighten prompts to require citations before any stakeholder-facing draft leaves the team.

FAQ

n8n required?
Optional; MCP-only teams can script writes from CI or IDE agents.

Can AI auto-resolve inbox items?
No—AI prepares evidence; humans resolve_decision_item.

Self-hosted n8n compliance?
Keeps flows in VPC; evaluate cloud n8n against policy.

Where is API docs?
See /api-docs for REST; MCP tool list in platform guidance.

Automate glue; debate forks; log bets

Let n8n and MCP kill copy-paste. Let simulate and boardroom govern choices that change dates, scope, and cash.

Try it now

Put these patterns on live project data—not slide decks.