The Admin Inbox — where context meets action.
A shared inbox where every feedback submission lands with its full captured context attached. Triage with your team. Export as a prompt. Let your coding agent read the report, open the issue, and draft the fix — with trust-aware routing keeping humans in the loop where it matters.
Everything in the inbox
Triage, context, integrate.
The trust model
Trust decides what's safe to automate.
Every submission gets a trust level based on how it arrived. Public widget submissions from anonymous users require human review before any agent can act. Signed-in reporters can be granted routing permissions. Agent and admin submissions can create issues immediately. You set the boundary.
Agent integration
Your coding agent reads the inbox.
Every feedback item has a <code>/context</code> endpoint that returns clean Markdown — message, console trace, failed network call, page state. Claude Code, Codex, or any agent with a scoped API key can read, comment, update status, and open GitHub or Linear issues. The MCP server exposes 10 tools for this workflow.
- Scoped API keys. Grant only what each agent needs:
feedback:read,feedback:comment,feedback:integrations, and more. - /context endpoint.
GET /api/feedback/:id/contextreturns Markdown the agent parses directly — no raw JSON wrangling. - MCP server.
quell mcpexposes 10 tools. Connect in Claude Code or any MCP-compatible agent. - Review-required PRs. Agent-opened pull requests always carry a review requirement. Merge stays with you.
FAQ
Inbox questions.
What is the Quell inbox?
The admin inbox is where all feedback submissions land — bugs, ideas, and praise from any project you've configured. Each item carries the full captured context: screenshot, console trace, network calls, and page state. You triage it, assign it, and either act on it yourself or hand it to an agent.
How does an agent read from the inbox?
Every feedback item has a <code>GET /api/feedback/:id/context</code> endpoint that returns a clean Markdown export — message, console trace, network failure, screenshot metadata, and page state. A coding agent like Claude Code or Codex reads this via a scoped API key or the <code>quell mcp</code> server.
What is the trust model and why does it matter?
Every submission gets a trust level: public, identified, reporter, agent, or admin. Trust decides what can be automated safely. Public feedback — from anonymous users via the widget — goes through human triage first. Agent and admin submissions can create GitHub or Linear issues immediately. You stay in control of the automation boundary.
Can an agent comment or update status in the inbox?
Yes. Scoped API keys grant fine-grained permissions: <code>feedback:read</code>, <code>feedback:comment</code>, <code>feedback:write</code> (status/assignee/due date), and <code>feedback:integrations</code> (create GitHub/Linear issues). The MCP server exposes 10 tools covering all these actions.
How is the inbox different from a standard helpdesk?
A helpdesk is built for conversations with customers. The Quell inbox is built for triaging technical feedback — bugs, regressions, usability problems. Every item carries structured developer context (console errors, network failures) that a coding agent can read without a human translating it. The trust model controls what gets auto-routed versus what needs human sign-off.
Can I filter or sort the inbox?
Yes. The inbox supports sort by newest, oldest, and priority, plus row selection for bulk actions — bulk status changes and bulk assign-me. Per-project filters let you focus on a single product. More filtering and search is on the roadmap.