Quickstart

Four settings stand between a new organization and its first governed agent run: a model provider key, at least one enabled MCP server, a policy that decides which tools run unattended, and the teammates who will use it. Do them in this order — each step unlocks the next.

Updated 30 July 2026 · 4 min read

Workspace access comes from your identity provider: the member who holds the admin membership role in WorkOS arrives in Lunen as an Admin, with the team, policy, secret, registry, and audit permissions that role carries. Everyone else arrives as a User. There is no separate bootstrap step inside Lunen.

Add a model provider API key

Open Settings → Secrets and paste an API key for at least one provider — OpenAI, Anthropic, or Google Gemini. Keys are write-only: once saved, the value can never be read back, only replaced or deleted. Until a key exists, no agent in the organization can run, and model pickers across the product stay empty.

Enable an MCP server

Open Settings → MCP Registry, find a server your team needs — Slack, Atlassian Rovo, Notion, GitHub, and dozens more — and switch on Enable … for this organization in its detail panel. Servers marked Needs setup list what they require (an OAuth client, server variables, or an install) before members can connect.

Decide the tool policy

Open Settings → Policies and set each of the server’s tools to Allow or Approve. Reads are usually safe to allow; anything that creates, updates, or sends should require approval until you have watched it run. The full model is on Policy & approvals.

Invite the team

Open Settings → Team and use + Invite to workspace. Each invitee gets a workspace access level — User runs agents and uses what is shared; Admin also manages team, roles, policies, secrets, and the registry. Members then connect their own accounts to enabled servers under Connected Services.

Watch one run end to end

Have an agent builder describe a small agent, run it once, and approve its first write from the run page. Then read the run in the audit log. Once you have seen the loop — plan, pause, approval, audit event — you know exactly what you are governing, and can widen the policy deliberately.

Note

Keys are not verified when you save them. A mistyped or unfunded key surfaces at first use, as a provider error on the run — not on the Secrets page.

Warning

Pending tool approvals expire after one hour and fail the run. Before you put agents on schedules, make sure someone is actually watching the approval queue.