Product

The support forum your AI can read, deployed per product.

Durable Q&A deflects repeat support and becomes the substrate AI answer engines cite. Every product in the suite runs its own community at community.<productDomain>, tended by bounded, auditable MCP agents: every capability is a named scope you granted, every risky move is a proposal, and every outcome is a ledger entry.

Create your communityRead the scope reference
5
grantable scopes
none
standing access
human
approval required
100%
actions traced
The lifecycle

One agent action, start to record.

  1. Grant

    You choose the scopes.

    A grant is a list of scopes on one community, nothing more. An agent with community:read can search and summarize but cannot write a single character.

  2. Propose

    The agent writes a diff, not a change.

    Every write-class tool returns a proposal: the current state on the left, the proposed state on the right, and the scope it used underneath.

  3. Approve

    A human decides.

    Approve, edit, or reject. Nothing in the proposal touches the forum until someone with the right role approves it.

  4. Record

    The ledger keeps the receipt.

    Actor, scope, trace identifier, and timestamp are written once and never rewritten. Both agent and human decisions are recorded the same way.

Step 02, as the reviewer sees it
acme.pubroo.linkIllustrative ledger entry
#getting-started

How should we welcome new members?

MKRAJDTS
Illustrative thread · not usage data
AgentAgent Moderator 7agent:moderator-7

Proposed adding a pinned setup checklist to this topic.

Currentpinned: nonesetup_guide: none
Proposedpinned: setup-checklistsetup_guide: "5 steps to first agent"
scope:community:write · trace:7f3a…c1 · 14:22 GMT
ApproveEditReject
The vocabulary

Five scopes. That is the whole vocabulary.

There is no catch-all permission and no implicit escalation. An agent holding community:read cannot reach a community:write tool, and the refusal is recorded like any other action.

  1. community:identity

    Read who the connected agent is acting as. Nothing about the forum itself.

  2. community:read

    Search topics, open threads, and summarize discussions. No writes of any kind.

  3. community:write

    Draft replies, propose tags, and flag content. Each one arrives as a proposal, not a change.

  4. community:moderate

    Read the moderation queue and triage review items. Destructive triage waits for confirmation.

  5. community:admin

    List members, change roles, create invitations, and update community settings.

Every tool and the scope it needs is listed in the MCP reference.

What you get besides the agent.

FORUM

A forum people actually want to use

Categories, tags, threaded topics, search, member profiles, invitations, and email notifications. The AI is additive, not a replacement for a working forum.

QUEUE

A moderation queue with a review trail

Flagged content collects in one queue. Every triage decision, human or agent, is a review item with a visible outcome.

LEDGER

An immutable audit ledger

Filter by actor, action, or target type. Team and above keep full audit ledger history for your own records.

BRAND

Your brand, your domain

Start on a pubroo.link subdomain, then move to your own domain with your own colors, logo, and sign-in page.

As of .

Read the contract before you sign it.

Free tier. Your domain. No credit card.

Create your community