Claude AgentMail connectorWhat Claude can do in your AgentMail account.
The Claude AgentMail connector exposes 24 tools. 10 read inboxes, threads, drafts and attachments, 13 create, send, update or delete, and 1 stays undocumented. Below: what Claude reads, what it can send or erase in your name, and what to check first.
Verified Trustpilot reviews · AI, automation & growth agency
Why connect an agent inbox to Claude
AgentMail is email built for AI agents: each agent gets its own address and can hold two-way conversations, not just fire off notifications. Once the connector is on, Claude can open a fresh inbox, read what came in, answer inside the thread and manage drafts, all from your chat. Several of those tools send real email or delete data, so the approval rules below matter.
Spin up a dedicated address for one project. create_inbox makes a new inbox, with an optional username, domain and display name, so a test or a supplier exchange stays out of your personal mailbox.
Answer in context. search_threads finds the conversation by full-text search, get_thread brings back its messages, and reply_to_message answers in the same thread, with everyone copied if you ask for reply-all.
Prepare now, send later. create_draft accepts a scheduled send time, and delete_draft cancels that send if plans change.
What it will not do: Claude does not watch the inbox. A new message does not make it react; it reads and writes only while it answers you. Several tools in AgentMail's own documentation (allow and block lists, provider sign-ups, organization switching) are not on the directory sheet; the comparison further down says which list to trust. For a mailbox connector from another publisher, see the Claude activecampaign connector page.
The vocabulary, quickly
Five words you will see on this page.
- Connector
- The link you create once between Claude and an account you own, so Claude can work in it while it answers you.
- Tool
- A named action the connector lets Claude perform. Claude decides which ones to call; the directory sheet lists every one by name.
- Authorization
- The service's sign-in and consent screen, where you hand over the access Claude will use. Once per person, and it can be revoked.
- Approval
- The confirmation Claude waits for before a step that changes your account, such as sending an email, shown in the chat when it matters.
- MCP
- The open standard connectors rely on: it is what lets an assistant like Claude call an outside service such as AgentMail.
Set up the Claude AgentMail connector
- 01
Look up AgentMail
In Claude's settings, open Customize, then Connectors, and find AgentMail. On a Team or Enterprise workspace, an Owner or Primary Owner has to enable it first; each member then signs in separately.
- 02
Sign in to the service
Hit Connect on its row and log in through the window the service opens, which is AgentMail's own page. If the link breaks later, Disconnect and reconnect, or remove the access from the service side.
- 03
Read the consent screen
Go through the authorization screen before accepting. It is the service's screen and it sets what the access covers, sometimes in broader terms than what Claude does daily; in Claude, approving each action is still the default.
The 24 tools: reading, writing, and one unknown
AgentMail gives Claude 24 tools: 10 that read your account, 13 that change something in it, and 1 no official source describes.
AgentMail's documentation describes every tool on the sheet but one, under the same names. One has no description anywhere. Names stay exactly as Claude shows them.
- 10 read
- 13 write
- 1 not documented
What Claude reads (10)
10 toolsTen tools that look at inboxes, messages, drafts and attachments without changing them.
get_attachment
Fetches one attachment from a thread with its metadata and a download link. For PDF and DOCX files, it also hands Claude the extracted text so it can read the document.
get_draft
Opens one draft by its ID with its content, status and any scheduled time. Claude can check what is queued ahead of that moment.
get_inbox
Returns the details of a single inbox from its ID. Claude uses it to confirm which address it is working in before doing anything else.
get_thread
Brings back a whole conversation by thread ID, messages included, so Claude reads the exchange in order before it answers or summarizes.
list_drafts
Shows the drafts of an inbox and can filter them by label, for instance the scheduled ones. You see at a glance what is waiting.
list_inboxes
Lists the email inboxes of your organization, page by page. AgentMail suggests asking Claude for exactly this to check the connection works.
list_messages
Returns the messages of one inbox, filterable by label, sender, recipient, subject or a before and after date. Good for sorting mail without a keyword.
list_threads
Gives the conversation threads of an inbox, with the same filters on labels, people, subject and dates. Claude gets the conversations rather than single emails.
search_messages
Runs a full-text search over the messages of an inbox and ranks hits by relevance. Claude finds a sentence even when you forget who wrote it.
search_threads
Does the same relevance-ranked full-text search, but returns whole threads, which suits questions about a conversation rather than a line.
What Claude creates, sends or deletes (13)
13 toolsThirteen tools that send email, change labels or settings, or delete data. No source describes a confirmation specific to any of them.
create_draft
Approval: see the ruleWrites a new draft email in an inbox, and can schedule it with a send time given in ISO 8601 format. The draft then waits until that moment.
create_inbox
Approval: see the ruleOpens a brand-new email address, optionally with a chosen username, domain, display name and metadata. Claude can then work in it with the other tools.
delete_draft
Approval: see the ruleRemoves a draft, and if that draft was scheduled, the planned send is cancelled with it. Nothing goes out afterwards.
delete_inbox
Approval: see the ruleErases an inbox using its ID. Of all the deletions this connector offers, it is the widest in scope.
delete_thread
Approval: see the ruleRemoves an entire conversation thread from an inbox, every message of that exchange at once.
forward_message
Approval: see the rulePasses an existing message on to new recipients. The email actually leaves your agent's address.
reply_to_message
Approval: see the ruleAnswers a message inside its own thread, and with reply-all it includes every original recipient. The reply goes out for real.
send_draft
Approval: see the ruleSends a prepared draft immediately. AgentMail turns it into a sent message and the draft itself is deleted.
send_message
Approval: see the ruleComposes and sends a new email from one of your inboxes to one or more recipients, without going through a draft.
update_draft
Approval: see the ruleEdits an existing draft and can move its scheduled send time to another moment.
update_inbox
Approval: see the ruleChanges an inbox's display name or its metadata. New metadata keys merge with the old ones, and a null value removes a key.
update_message
Approval: see the ruleAdds labels to a single message or takes them off, so Claude can sort mail as it reads it.
update_thread
Approval: see the rulePuts labels on a whole thread or removes them, for every message of the conversation at once.
Not documented (1)
1 toolThe directory publishes this name and nothing else. No official source says what it does, so we describe nothing.
auth_me
The directory sheet lists this tool by name, and no official source, neither AgentMail's documentation nor any Claude page, describes what it does. We do not guess.
What Claude asks before sending or deleting
This connector can send real email and delete inboxes, threads and drafts. That is where the approval rules earn their keep.
By default, Claude stops and asks for your approval before each action it takes on an account for you. On Team and Enterprise, owners decide whether members may let some actions run without a prompt every time, and they can restrict a connector for the whole organization, for example allowing reading while blocking sending; members cannot override that. Claude also works with the identity of the person who connected: AgentMail says the server uses your console identity for every call, so it reaches what that identity reaches.
Plans: the honest answer
Plan availability for AgentMail in Claude is not published anywhere. Not one of the 819 directory sheets shows plans.
The rule Anthropic does publish: remote connectors are open to all users on Claude, Cowork, Claude Desktop and mobile, and on Team and Enterprise an Owner or Primary Owner enables a connector before members can connect. For this one's current state, check its sheet in the official directory. Only AgentMail documents this connector; no Claude help page covers it.
What the AgentMail connector does not do
A connector is not an automation. Claude calls these tools while it answers you; an incoming email does not wake it up or trigger a reply.
The partner badge in the directory is not a security audit, and Anthropic states on every sheet that it neither picks the tools a publisher exposes nor guarantees they behave as announced. Connect only what comes from a publisher you trust. AgentMail's documentation also describes a wider server than the sheet shows, which is why the two disagree below. For other connectors on the site, start from the Integrations hub.
How many tools does AgentMail give Claude?
- Claude directory sheet ↗2026-10-0124 tools, including auth_me, and none for allow and block lists, providers or organizations.
- AgentMail documentation (MCP) ↗2026-10-01A larger catalog grouped by resource, adding allow and block lists, AgentID, providers and organization selection.
What this page followsThe sheet is what Claude displays; AgentMail says the live catalog comes from its server. Neither settles it for your account: check the tool list in Claude once connected.
Need help connecting AgentMail to Claude?
A person reads every message.