AgentMail
Give an AI agent its own email inbox: create inboxes, read threads and send mail to anyone.
Send transactional email, run broadcasts to a contact list, manage domains and templates, and read inbound mail.
Opens Agent Studio, where connecting is one click. The connector URL below works in any MCP client.
Paste it into any MCP client. Setup docs
The Resend MCP server connects Resend's email platform to Claude, ChatGPT, Cursor and any MCP-compatible agent. It sends transactional email, builds and sends marketing broadcasts to a contact list, manages sender domains and templates, and reads inbound mail — all in conversation. Sign-in is OAuth, and a verified domain is required before anything sends.
Verified connector
Listed by Anthropic as a partner connector in its Connectors Directory.
Connection checked by Agentman on .
Anthropic states this reflects the level of review a connector received, not a security audit.
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
send-email reaches up to 50 recipients per call per Resend's source. send-broadcast reaches an entire contact segment. Resend documents no server-side confirmation step for either, and neither tool carries a safety annotation your client could act on. The only thing standing between a misread instruction and a delivered email is your MCP client's approval prompt. Configure it before connecting a production account.remove-* tool plus revoke-oauth-grant and batch-remove-suppressions — carry an instruction inside the tool description telling the model it must double-check with the user, warn that the action is irreversible, and may only proceed on explicit confirmation. That is a model instruction, not an enforced gate. It is absent from send-email, send-batch-emails and send-broadcast. Resend guarded every deletion this way and no send.sending_access-scoped API key is the nearest approximation, and it still sends.create-api-key mints a credential, and the reach boundary is undocumented. The tool takes a permission of full_access or sending_access; Resend describes full_access as able to create, delete, get and update any resource, and sending_access as restricted to sending, with an optional domain_id narrowing it to one domain. Resend's source instructs the model to display the token to the user because it is shown only once. What Resend does not document is whether an OAuth session can mint a key broader than itself. Resend's authorization server advertises the scopes full_access and emails:send, so a narrow grant is expressible — but Resend publishes no rule connecting a grant's scope to the permission an issued key may carry. We did not test it, and we make no claim either way. Treat it as an open question and prefer revoking rather than creating keys through an agent.scopes_supported field at all. Its authorization server separately advertises two — full_access and emails:send — so a send-only grant exists at the platform level. Because the MCP resource declines to enumerate scopes, we cannot say which of the two the connector requests, and we do not claim it requests either. Check what the OAuth screen actually asks for before approving.403 and a validation_error. Both sent and received email count toward the quota, and so do test sends to resend.dev addresses.cancel-broadcast and a six-tool suppression list surface. A page or client sized on the 85-name listing will not know the connector can add and remove suppressions — the mechanism that determines who never hears from the account again.to field.Yes. Resend states that before you can send or receive emails you must have a verified domain associated with your account, added and confirmed through DNS records. Resend says domains often verify within 15 minutes of the DNS records being added, but that propagation can take up to 72 hours. Resend recommends sending from a subdomain rather than a root domain.
Not for sending. Resend's open-source server writes a confirmation instruction into the descriptions of all 14 of its removal and revocation tools, telling the model it must double-check with the user first. The send tools carry no such instruction and no destructive safety annotation, so whether a send is confirmed depends entirely on your MCP client's approval settings.
Resend's documentation does not say. The create-api-key tool accepts a permission of full_access or sending_access, and Resend describes full_access as able to create, delete, get and update any resource. Resend publishes no rule about how a key's permission relates to the OAuth grant that created it, so treat that boundary as undocumented.
Send to Resend's reserved resend.dev addresses. Resend documents delivered@resend.dev, bounced@resend.dev, complained@resend.dev and suppressed@resend.dev, each simulating one delivery event, and supports labels after a plus sign. Resend notes test emails still count against your account's sending quota, and blocks example.com and test.com addresses with a 422 error.
Resend documents a default maximum of 10 requests per second per team, applied across every API key on the team, and says it can be raised for trusted senders on request. Separately, the Free plan allows 3,000 emails per month capped at 100 per day. Exceeding a quota returns a 429 with daily_quota_exceeded or monthly_quota_exceeded.
Any plan, including Free. Resend's documentation names no plan requirement for the MCP server itself and gates nothing on tier. What your plan limits is volume: the Free tier covers 3,000 emails per month and 1,000 marketing contacts, and Resend says you cannot send a broadcast once you exceed your contact quota.
Yes. Resend publishes the server as the resend-mcp package on NPM and states the hosted server runs the same open-source code. The local server takes a Resend API key rather than OAuth and supports stdio and HTTP transports. Resend lists creating an API key and verifying a domain as its prerequisites.
.md variant Resend serves; ~1,600 words). Source for the endpoint, client setup, the local server, the API-key-as-Bearer alternative and the 14-group tool summary. · retrieved 2026-08-18llms.txt — and (retrieved 2026-08-18, ~48KB index). The machine-readable page index that located every doc page cited here. Resend serves clean Markdown by appending .md to any docs URL. · retrieved 2026-08-18daily_quota_exceeded / monthly_quota_exceeded errors. · retrieved 2026-08-18resend.dev simulation addresses, label support, the quota note and the 422 on example.com. · retrieved 2026-08-18full_access / sending_access permission enum and the optional domain restriction. · retrieved 2026-08-18resend-mcp server source (package version 2.13.0, main branch commit dated 2026-08-11, retrieved 2026-08-18). Source for the 92-tool count, the 38 read-only markings, the zero destructive markings, the confirmation instructions inside five tool descriptions, and the 50-recipient cap on send-email. Resend's own MCP page states the hosted server runs this code. · retrieved 2026-08-11/mcp endpoint, the OAuth configuration and a 12-group tool taxonomy. · retrieved 2026-08-18initialize against https://mcp.resend.com/ — HTTP 401 with WWW-Authenticate: Bearer resource_metadata="https://mcp.resend.com/.well-known/oauth-protected-resource", retrieved 2026-08-18. This authentication challenge is what establishes the OAuth posture. The same probe against https://mcp.resend.com/mcp returned the identical 401. · retrieved 2026-08-18https://api.resend.com as the authorization server and publishes no scopes_supported field, so we make no claim about which scope the connector requests. · retrieved 2026-08-18scopes_supported of full_access and emails:send, a registration_endpoint (dynamic client registration), and code_challenge_methods_supported of S256 (PKCE). · retrieved 2026-08-18partner tier, endpoint, transport and categories above are read from our committed directory snapshot dated 2026-08-16; the directory page itself blocks automated fetches, so we did not retrieve it directly. · retrieved 2026-08-16resend.com/robots.txt allows this path and carries Content-Signal: ai-train=yes, search=yes, ai-input=yes (retrieved 2026-08-18) — an explicit opt-in to AI synthesis, which is the basis on which this page restates Resend's documentation. · retrieved 2026-08-18Connect Resend once and your agents call these tools on their own: on a schedule, in a workflow, with nobody at the keyboard.