Unsendbox logo
Get started
← RETURN TO ROOT
Connected

User guide

Your daily workflow should start with MCP in ChatGPT/AI chat, then use the UI for verification and exceptions.

Start with getting started

Recommended workflow order

Use this order for predictable setup and lower operations drift.

Step 1 — Setup email (test)
Complete domain + address onboarding, then run a controlled test send.
Step 2 — Send and receive
Use `send_email`, `list_threads`, and `reply_to_thread` for daily operations.
Step 3 — Forwarding and routing
Add forwarding only after base flow is stable. For Workspace-first teams, keep Google MX/IMAP and forward explicit AI identities.
Step 4 — Safety layer
Protect routing with `set_mirror_rule` and review exceptions in the UI.
Step 5 — Verify in UI
Cross-check thread and delivery state from dashboard before volume traffic.
Step 6 — Opencode control plane
Use OpenCode+Codex to own the MCP interaction layer and keep UI for exception handling.

Unsendbox MCP surface (current API)

These tools are available in Unsendbox MCP and are mirrored in this guide.

Domains
`connect_domain` · `disconnect_domain` · `list_domains` · `get_domain_health`
Addresses
`create_address` · `delete_address` · `list_addresses` · `toggle_address`
Outbound
`send_email` · `reply_to_thread`
Threads
`list_threads` · `get_thread` · `search_threads`
Mirror rules
`set_mirror_rule` · `list_mirror_rules` · `delete_mirror_rule` · `toggle_mirror_rule`
MCP prompts
`getting-started` · `email-management` · `deliverability-setup`
Opencode layer
`opencode.json` project config, Unsendbox MCP server registration, local hooks, and agent prompts.

Workspace-first + Unsendbox strategy

Use this when you want Google for mailbox UX and Unsendbox for AI-driven control.

Primary mailbox
Google Workspace handles IMAP, labels, mobile, and end-user mailbox behavior.
Primary automation plane
Unsendbox owns route creation, mirror behavior, and suppression checks for outbound safety.
Boundary guardrail
Only forward selected identities or tags to Unsendbox to avoid loops and ensure review points stay clear.

MCP-first workflow (recommended daily driver)

Use your AI chat with the Unsendbox MCP as the primary way to operate mail operations.

Why this order
Chat gives fast, repeatable commands and keeps domain onboarding, sender setup, and sends in one auditable history.
Minimum first flow
  • Start a ChatGPT/Claude/OpenAI chat session connected to Unsendbox MCP.
  • Run the Unsendbox onboarding prompt.
  • Use `connect_domain` for DNS verification.
  • Create sender addresses with `create_address` once domain is connected.
  • Draft and send via `send_email` or `reply_to_thread`.
Best for
Teams that want one source of truth for operational commands and minimal UI drift during execution.

MCP runbook for day-one setup

Follow this in AI chat before opening dashboards.

  1. 01Open the Unsendbox MCP prompt and run the recommended onboarding sequence.
  2. 02Call `connect_domain` with your zone name and complete DNS records in Cloudflare.
  3. 03If your team is Workspace-first, confirm whether Google or Unsendbox should be authoritative for domain MX.
  4. 04Run `get_domain_health` and ensure routing/DNS checks are healthy.
  5. 05Create dedicated senders for each agent role with `create_address`.
  6. 06If using Workspace-first routing, create aliases/forward targets for explicit AI identities.
  7. 07Use `send_email` or `reply_to_thread` for first test traffic and verify outcomes.

OpenCode onboarding path (before daily operations)

Do this once per device/workstation before using ChatGPT/Codex as your daily driver.

  1. 01Create the repo Opencode config from `/docs/configuration/opencode-setup`.
  2. 02Enable `opencode-openai-codex-auth` and set the Codex/ChatGPT token in env.
  3. 03Point the Unsendbox MCP URL to the org endpoint your token can call.
  4. 04Set `list_threads` and `get_domain_health` as first health checks in every session.
  5. 05Run a test automation loop (send, receive, mirror rule, reply) from Opencode before production.

Operational cadence in chat

Daily operations should stay in chat, especially for repetitive actions.

  1. 01Create/update mirror rules for critical human review paths before large campaigns.
  2. 02Use `list_threads`/`search_threads` to inspect replies before composing follow-ups.
  3. 03Keep deliverability checks in your pre-send loop and capture failures with IDs.
  4. 04When blocked, ask the assistant for next action and recovery sequence before retrying.

UI counterpart section

Use the UI for monitoring, manual exceptions, and explicit approvals.

Dashboard
Monitor overall status, recent activity, and workflow completion at a glance.
Domains and addresses
Review DNS and routing health, onboard additional domains, and confirm sender identity status.
Inbox and threads
Inspect incoming/outgoing messages visually when human review is required.
Rules and safety
Apply mirror and suppression settings for policy-level control instead of chat-only changes.
Forwarding validation
Use domain and address screens to verify Workspace-forwarded identities and avoid loops.

Use both layers together

Keep chat and UI in sync instead of choosing one permanently.

  1. 01Run normal operations in chat for consistency, then validate outcomes in the UI.
  2. 02Use UI rule screens if you need to inspect or fine-tune routing exceptions.
  3. 03Keep your MCP history as the source of truth; use UI for visibility and exception handling.
  4. 04If there is any mismatch, treat chat commands as current and reconcile in UI after the fact.