Common AI agent patterns teams run against Onboard with MCP—Chief of Staff, setup, onboarding, review, reporting, opportunity, and risk.

Focused agents, one MCP connection

MCP

Several narrow assistants beat one generic chat: each pattern below maps to tools your client can call with the same Onboard credentials.

Many teams deploy several focused agents instead of one generic assistant. Each agent has a narrow job, clear inputs, and human checkpoints. Together they mirror how a mature onboarding or CS org already divides work.

These patterns assume you have followed MCP setup and Connecting to Onboard MCP, and that you follow Security & permissions (MCP) and the Security checklist.

For implementation teams using Cursor or Claude in the onboard-rest-api repo: each persona has a matching agent skill under .cursor/skills/onboard-agent-* (mirrored under .agents/skills/ and .claude/skills/) with an MCP tool playbook (browse → explain → run_api_request). The canonical technical reference is docs/api/mcp-agent-tool-playbook.md at the repository root (next to onboard_mcp/). Customer language: say Project; OpenAPI still uses tag Map for filters—see Introduction to MCP.

Chief of Staff

Often the first agent to stand up. It reviews overall progress across accounts or programs and produces start-of-week and end-of-week briefs: what moved, what stalled, and what needs a human decision next.

Use it when leadership wants a consistent narrative without pulling manual reports every Monday and Friday.

Set-up agent

This agent helps configure Onboard for a new program or customer: tasks, templates, branding, users and roles, custom fields, custom forms, KPI dashboards, meeting insights, bots, context filters, and other features as your team adopts them over time.

A critical step is discovery: the agent should run short interviews (or structured questionnaires) with the customer about process and goals before it recommends or applies configuration. Humans approve material changes.

Onboarding agent

A core operational agent. It uses Onboard regularly with scheduled cadence—for example morning and evening summaries with clear to-dos: nudges for late tasks, deadlines at risk, and open customer threads. It also handles ad hoc work such as drafting replies to messages (for human send) and surfacing priority notifications.

This is what many stakeholders will experience as “AI automation” day to day. Keep customer-facing sends and policy changes under human control.

Review agent

Focused on health and accountability. It reads KPI dashboards and related signals to explain how projects are progressing, which workstreams are behind and need attention, and where coaching or training might help the team—not individual blame, but capacity and skill gaps surfaced early.

Reporting agent

Answers ad hoc reporting questions by gathering data from Onboard (and, where you connect them, other systems). Examples: pipeline snapshots, cohort comparisons, or “documents uploaded by customers ABC and DEF” for a review meeting.

Treat exports and document lists as sensitive; scope access to the same rules as your analytics and privacy policies.

Opportunity agent

Runs a daily pass looking for upsell or expansion signals—usage patterns, completed milestones that open the next conversation, or product interest captured in notes. It works best when Meeting insights and context filters are configured so signals are trustworthy.

Opportunities should feed sales or CS playbooks; humans own outreach and commercial terms.

Risk agent

Also daily, oriented toward risk: churn indicators, sentiment shifts, or accounts that look angry or disengaged. Like the opportunity agent, it benefits from Meeting insights and context filters so noise stays low.

Route severe risk to your escalation process; the agent recommends, it does not replace human judgment on renewals or crisis response.

How these fit together

Chief of Staff sets the weekly rhythm.
Set-up runs once (or per segment) with discovery first.
Onboarding carries daily execution.
Review, Reporting, Opportunity, and Risk specialize on cadence and question type.

For narrative playbooks (leadership review, launches, follow-ups), see Example workflows. For day-to-day prompt patterns, see AI workflow examples.

MCP use cases — AI agents