Let your AI agentrun your email
Claude, ChatGPT, Cursor and Codex connect over MCP. Your agent sends, edits and reports with only the access you approve.
From prompt to first sendin three steps
Add one URL to your agent, approve what it can touch, then ask for what you need in plain words.
The same toolsthe dashboard uses.
Twenty-seven tools across eight areas. Each one needs a scope you approved, and each says whether it only reads or changes something.
Emails4 tools
emails:reademails:write- send_email
- get_email
- list_emails
- get_stats
Contacts9 tools
contacts:readcontacts:write- list_contacts
- get_contact
- create_contact
- update_contact
- set_contact_topics
- delete_contact
- list_segments
- list_topics
- list_contact_properties
Campaigns7 tools
campaigns:readcampaigns:write- list_campaigns
- get_campaign
- create_campaign
- update_campaign
- preview_campaign
- list_campaign_versions
- restore_campaign_version
Templates8 tools
templates:readtemplates:write- list_templates
- get_template
- create_template
- update_template
- publish_template
- list_template_versions
- restore_template_version
- preview_template
Domains2 tools
domains:read- list_domains
- get_domain
Automations2 tools
automations:readautomations:write- list_automations
- set_automation_status
Webhooks1 tool
webhooks:read- list_webhooks
Help1 tool
templates:read- guide
Watch it editnext to your team.
Your agent edits the live draft, so each change shows in the editor under its name. It checks the version it read first, so it never writes over a teammate.
- Find and replace, insert, move or delete blocks
- Every change applies, or none of them do
- Nothing sends until someone schedules it
Only the accessyou give it.
You approve each agent once, for one workspace. You see what it did, and you can cut it off at any time.
- Choose the ScopesRead or write, area by area, for the one workspace you pick.
- Revoke Any ConnectionEvery connected agent is listed. Revoke one and its tokens stop working.
- See Every Call in LogsEach tool call shows with MCP as the source, with its request and response.
Safe to retry,safe to run twice.
Agents time out and try again. A retried send returns the first email, and an edit made from an old copy is refused.
An assistantinside Inboxjot.
Ask for what you need in the editor. It works next to your team and waits for your yes before anything that sends, deletes or suppresses.
- Edit Campaigns LiveAsk for a shorter intro or a new section, and watch the draft change next to your team.
- Approve Before It ActsAnything that sends, deletes or suppresses waits for your yes, with a card that shows what will change.
- Undo Puts It BackEvery change it makes is in the editor’s history, like a teammate’s.
And everything elseyour agent needs.
Scoped Access
Read or write for each of eight areas, approved for one workspace.
OAuth Sign-In
Agents sign in through Inboxjot. No API key pasted into a chat.
Revoke Anytime
Cut off any agent, and it stops working right away.
Every Call Logged
Tool calls show in Logs with MCP as the source.
Audit Log
Each agent you connect or revoke is in the audit log.
Docs for Agents
llms.txt, llms-full.txt and a Markdown copy of every docs page.
Nine Clients
Setup steps for Claude Code, Claude, ChatGPT, Cursor, Codex, Pi, VS Code, GitHub Copilot and Windsurf.
Safe Sends
Agent sends count toward your daily limit, and suppressed addresses are refused.
Questionsabout AI agents
More fromInboxjot
Your agent’s first sendis one prompt away
We review every new account, usually within a day.