A Gmail assistant that can search an inbox still cannot send your SaaS app’s password-reset messages through its delivery system. That difference is central when you choose an email MCP server: the right setup can help an AI client work with an inbox, trigger application email, or support those jobs through separate integrations. This guide explains the options, setup decisions, and safeguards so you can match the connection to the work.

An email MCP server exposes email-related actions to an AI client through the Model Context Protocol. A host manages MCP clients, each client communicates with one server, and that server makes its tools, resources, or prompts available. Some servers connect to inboxes for reading and organizing messages; others connect AI workflows to application email infrastructure for sending. MCP standardizes the interaction, not the email capabilities: the server and its provider determine what the agent can do.

What is an email MCP server?

The Model Context Protocol (MCP) is a protocol for connecting AI applications to external capabilities. For email, a server might expose tools for searching a mailbox or sending a message through an email service. The AI application acts as the host, creates an MCP client to communicate with a server, and decides how to use the results. MCP uses JSON-RPC for this communication.

The parts have distinct responsibilities:

  • Host: The AI application coordinates the model, creates and manages MCP clients, and handles connection permissions and user consent.
  • Client: A component managed by the host that exchanges protocol messages with one server. A host can manage several clients, but each client communicates with exactly one server.
  • Server: A local process or remote service that exposes capabilities. Its tools are the operations the client can request; resources and prompts can provide other kinds of context.

For example, a person might ask an AI assistant to find a receipt email. If the connected server offers a mailbox search tool, the client sends a request to that server, which searches through its configured provider connection and returns matching information. The model can then summarize the result. The server’s available tools determine whether it can also draft, send, or organize messages.

MCP is not a replacement for email protocols. IMAP4rev2, specified in RFC 9051, defines how clients access and manipulate messages and mailboxes on a server. MCP gives AI applications a standard way to request actions from a connected server, which may use IMAP, a provider API, or an email platform behind the scenes.

The current MCP specification is version 2026-07-28, released on July 28, 2026, according to the MCP announcement. Its architecture describes a host managing client instances and servers exposing capabilities. A client can discover a server’s capabilities before making another request, but capability discovery is optional (MCP architecture).

What can an email MCP server do?

An MCP server can only perform actions its implementation exposes and its provider permissions allow. For inbox work, those actions often map to operations supported by IMAP4rev2, such as searching mailboxes, retrieving selected message content, and manipulating folders or message flags. IMAP4rev2 also lets clients check for new messages and resynchronize offline mailboxes.

Common mailbox tasks include:

  • Read: Retrieve a message or selected parts, such as its subject and body.
  • Search: Find messages by criteria and fetch matching results. IMAP4rev2 supports searching and selectively fetching message attributes, text, and portions of messages.
  • Draft or send: Prepare a message or reply, then send it if the server exposes a sending action and the account permits it.
  • Organize: Manage folders or message flags, if the relevant server tools support those operations.
  • Monitor: Surface new messages or changes when the server offers a watch or notification capability.

These actions help an agent work with correspondence already in an inbox. A support workflow might search for a customer’s latest message, summarize the thread, and prepare a response for a person to review. The specific actions vary: access to search does not imply access to sending, deletion, or monitoring.

Application email has a different purpose. A SaaS product may send a verification message after signup, a receipt after payment, or a password reset after a user requests one. Those messages originate from an application event, not from an agent searching a personal inbox. Campaigns and automated workflows add audience-based and event-triggered sending to that delivery use case.

A server that can send one message may not support campaigns or automation. Likewise, a mailbox connection does not automatically provide application delivery infrastructure. Read the server’s tool list and the email provider’s feature descriptions as separate evidence: the word “email” does not establish which of these jobs it supports.

Mailbox connections vs. transactional email integrations

A mailbox connection lets an AI client work with a person’s or team’s existing messages. A transactional email integration helps an application send messages in response to product events. Teams sometimes need both, but the two connections should not be treated as interchangeable.

IMAP is designed for accessing and manipulating messages and mailboxes. SMTP is designed for Internet email transport and delivery, not for reading an inbox. The SMTP specification, RFC 5321, describes SMTP as a transport and delivery protocol and notes that a separate mail-submission protocol is generally preferred to direct SMTP use. So an IMAP SMTP MCP server may combine inbox access through IMAP with outgoing mail through SMTP; SMTP alone does not provide mailbox search.

A product’s transactional email setup typically uses an API or SMTP relay to submit application messages. The application decides when to send a receipt or account notification, while the email service handles delivery. An MCP integration can make some of that service’s capabilities available to an AI client, but it should not be assumed to grant access to customer inboxes.

Use mailbox access when a task depends on incoming messages, searching, or working on replies. Use transactional infrastructure when software must send product-triggered messages. Add campaigns or event-triggered workflows when the requirement extends to audience communication or automated sequences. For example, a SaaS support agent might use inbox access to draft a reply while the product sends account-verification messages through a separate delivery service.

Email MCP server options to consider

The main options fall into three groups: general IMAP and SMTP servers, Google Workspace projects that include Gmail capabilities, and transactional email services with an MCP listing. Compare their documented actions, provider fit, setup needs, and intended use rather than treating them as interchangeable products.

OptionDocumented capabilities and providersSetup and fit
General IMAP/SMTP serverThe IMAP and SMTP protocols can support mailbox operations and outgoing mail respectively. Which providers, accounts, and MCP tools are supported depends on the individual server.Suitable for working with existing mailboxes when the server supports the provider’s protocols and authentication. SMTP is for sending; IMAP is the mailbox-access part.
Google Workspace CLI (gws)The Google Workspace CLI repository lists Gmail helpers for sending, replying, reply-all, forwarding, unread-inbox triage, and watching for new messages. It also includes an MCP server. Its commands are built dynamically from Google Discovery Service.A Google Workspace-oriented option for Gmail and other Workspace services. The repository says the project is not an officially supported Google product, and it does not provide a definitive list of exposed MCP tools.
Google MCP repositoryThe Google MCP repository lists an open-source Google Workspace server that includes Gmail and is described as a Gemini CLI extension.The repository describes the project as a demonstration and says it is not an officially supported Google product. Its published description does not establish a complete Gmail tool inventory or exact authorization requirements.
MailerSendMailerSend lists an MCP server alongside its Email API and SMTP relay. Its pricing page also lists SMTP and API documentation, 7 SDKs, and an OpenAPI collection.A transactional email option for application delivery. The pricing page does not define the MCP tool list or say that the integration provides general-purpose inbox access.
MailtrapMailtrap’s pricing page separates Email Live Sending from Email Sandbox, which provides test email capacity. Its live sending plans include sending limits and email logs.Consider it when evaluating live application delivery or a dedicated email testing workflow. The pricing information does not establish a specific MCP server or set of MCP actions.
SendGridSendGrid describes its Email API as a way to send at scale and provides an SMTP service. Its API supports transactional sends and a Marketing Campaigns API for lists and campaigns.A sending and campaign platform with API and SMTP options. Its integration directory does not identify an MCP server or list MCP-supported actions.

MailerSend’s pricing page lists a free tier with 500 emails per month, Email API, and SMTP relay. It also lists a Hobby plan at $5.60 per month when billed yearly ($67.20 yearly), or $7.00 per month, with 5,000 emails per month. The page lists Starter at 50,000 emails per month and Professional at 50,000 emails per month, but does not state legible current monthly prices for those two tiers.

Mailtrap’s pricing page lists Email Live Sending Free at $0 per month for up to 4,000 emails, with a 150-email daily limit. Its Email Sandbox Free tier is separate: it includes up to 50 test emails. That distinction matters when comparing a service used to test messages with one used to send to recipients.

SendGrid provides a concrete example of API-based sending: its support documentation shows a POST https://api.sendgrid.com/v3/mail/send request that needs at least a recipient, sender, subject, and content. The sender address must belong to a domain authenticated on the account. SendGrid also documents marketing campaign and transactional template APIs, but its integration directory does not identify an MCP server.

How to choose the right connection

Start with the task the agent must complete, then narrow options by provider compatibility and control. A server with a broad tool catalog is not automatically a better match; it may expose actions your workflow does not need.

  • Inbox or application delivery: Choose mailbox tools for reading and organizing received messages. Choose an API or SMTP-based service for product-triggered sending. If you need both, plan two integrations unless one option documents both capabilities.
  • Provider compatibility: Match the integration to the actual account and its authentication method. Google’s gws repository lists Gmail actions such as +send, +reply, +forward, +triage, and +watch; do not assume another provider offers the same commands.
  • Account scope: If an agency or support team uses several inboxes, confirm how the server handles multiple accounts and separates access. A general protocol label does not establish multi-account support.
  • Campaigns and workflows: Confirm that the email platform offers campaign management or event-triggered workflows if those are requirements. A single-message sending tool is not proof of either capability.
  • Setup and permissions: Compare the required provider configuration, authentication, and client support. Limit the first connection to the actions needed for its initial task, then expand only when a workflow requires more.
  • Operational fit: Consider message logs, testing options, contact storage, monthly volume, and what happens at a sending limit. These affect the application’s behavior after deployment, not just initial setup.

A practical rule is to choose the narrowest integration that completes the job. An agent summarizing unread Gmail needs mailbox-reading capabilities; a SaaS app sending account alerts needs application delivery. If support staff also reply to customers, keep the inbox connection and product-sending path distinct so a credential for one is not mistaken for authorization to use the other.

How to connect a server to an AI client

Connection steps differ among servers, providers, and AI clients, but the sequence is consistent: select the job, prepare provider access, configure the client, and test a low-impact operation. Our setup guide for Gmail, SMTP, and Volanea covers those connection categories in more detail.

  1. Choose the provider and server. Decide whether the client needs mailbox access or application sending. Confirm the server documents the provider, actions, and transport you need.
  2. Prepare credentials or authorization. Enable any required provider API and create the credentials requested by that server. For OAuth, understand the requested access before authorizing it. For protocol credentials, use a protected secrets store rather than placing a password or key in a prompt or checked-in configuration file.
  3. Configure the MCP client. Add the server using the client’s supported connection method and provide authentication through its documented mechanism. MCP servers can run as local processes or remote services, so configuration depends on the server’s transport.
  4. Limit initial access. Begin with the smallest permission set that supports the first task. For an inbox summary, start with read and search capabilities rather than enabling send or delete actions.
  5. Test a safe operation. Try a bounded task, such as searching for a known test message or listing available capabilities. Confirm the result in the provider interface before using the connection for sending or changes to real mail.

For an application-sending setup, a safe first test is a message to an address controlled by your team, using the provider’s test mode if available. Check the rendered message and delivery logs, then confirm the intended sender identity and application event. A successful connection only demonstrates that one operation worked; test each action your deployed workflow depends on.

Security risks and safeguards

An email connection can expose private correspondence or allow consequential actions. Treat it as an application integration with defined permissions, credentials, and review points—not as an ordinary chat feature. The host and client architecture gives the application a role in managing connection permissions and user consent, but teams still need to decide what access is appropriate for each workflow (MCP architecture).

Use OAuth or another provider-supported authentication method and protect credentials in a secrets manager. Do not place tokens, passwords, or API keys in prompts, source control, or shared logs. For IMAP, RFC 9051 says LOGIN does not provide privacy protection or integrity checking and clients must not use it on an unsecured network. The specification requires servers to disable LOGIN unless the connection is protected, and clients not to use it when LOGINDISABLED is advertised.

SMTP connections also depend on correct TLS behavior. The STARTTLS extension lets a client and server negotiate TLS; after a server replies 220 to STARTTLS, the client must begin TLS negotiation before sending any other SMTP commands. RFC 3207 says a client should abort if it cannot start the TLS handshake, and should issue QUIT if the resulting TLS connection does not provide sufficient privacy or authentication. RFC 4954 requires SMTP AUTH implementations to support the PLAIN SASL mechanism over TLS and prohibits AUTH during a mail transaction or a second AUTH after successful authentication in that session.

Require human review before an agent sends, deletes, or materially changes messages, especially during initial deployment. Keep logs useful for investigating tool use, but restrict who can read them and avoid retaining message bodies or secrets without a clear need. Test prompts and workflows with untrusted email content; an email is input data, not authorization to override the user’s task or the application’s access rules.

For a practical review of credential scope, approvals, and operating boundaries, see our email MCP permissions guide and secure setup checklist. Those safeguards apply whether a connection is used for an inbox or a delivery service, though the specific risks differ by the actions it exposes.

Where Volanea fits for developer email workflows

Volanea is for teams that want transactional email, campaigns, and automated workflows for their product, rather than an assistant that operates a personal inbox. We send through Amazon SES and provide transactional sending, campaigns, and event-triggered automation through one API. Our MCP server is available on every plan and includes 23 tools; the tool names and capabilities are visible through the connected server, so an agent can work with the actions it exposes without treating the integration as Gmail or Outlook access.

A developer workflow might look like this: ask a coding agent to prepare an account-verification email for a signup flow. The agent connects to the Volanea project, inspects the available MCP tools, and uses the appropriate exposed actions for the email task. The developer reviews the generated template and application code, configures the app to call the transactional API after signup, and sends a test to a controlled address before enabling the flow for users. The email’s from field can be a bare address such as updates@example.com or a display-name form such as Product Updates <updates@example.com>.

For an SMTP-based application, the login username is volanea and the password is the project’s secret API key. A sk_test_ key can render and log messages but never delivers them, which makes it useful for checking application behavior without sending real mail. These details are specific to Volanea’s SMTP relay; an IMAP inbox connection has a different purpose and credentials.

Every feature ships on every Volanea plan; plans differ only by monthly email volume and sending-domain count. The free plan includes 1,000 emails every month, unlimited contacts, the full set of 23 MCP tools, and no-card-required signup. The paid tiers include Starter at $5 per month for 7,500 emails, Growth at $20 per month for 50,000 emails, and Scale at $149 per month for 200,000 emails. The published pricing page also presents annual-billing figures: Volanea Starter for 12 months at the listed monthly rate is 4 × 12 = $48; Volanea Growth for 12 months at the listed monthly rate is 16 × 12 = $192.

There are no overage charges. When a project reaches its monthly sending limit, sending pauses until the limit resets or the plan changes. Volanea sends from shared Amazon SES infrastructure and does not offer dedicated IPs, so a dedicated sending IP requirement rules it out. A BAA is available on request via hello@volanea.com; Volanea has not obtained SOC 2 or ISO 27001 certifications.

A concrete planning example: a product sending roughly 6,000 account notifications monthly fits within Starter’s 7,500-email volume, while a team expecting 30,000 monthly messages would need a tier with at least that volume, such as Growth. Contacts are stored for $0, so choosing a tier is about sending volume and sending-domain count, not the number of stored contacts. The published plans have no overage billing, so teams should account for the pause at the limit when setting application alerts and operational expectations.

Choose Volanea when the job is application email plus campaigns or event-triggered workflows, and you want an MCP connection for coding-agent tasks alongside a REST API or SMTP relay. Choose an inbox-focused server when the agent needs to find, read, or organize existing messages. Our MCP email sending tools and costs comparison helps compare provider options for the delivery side.

FAQ

Is there an MCP server for Gmail?

Yes. The Google Workspace CLI repository includes an MCP server and lists Gmail helpers for sending, replying, forwarding, triage, and watching for new messages. The repository says the project is not an officially supported Google product. The separate Google MCP repository also lists a Workspace server that includes Gmail and describes it as a demonstration project.

Is there an MCP server for Outlook?

The options described in this guide do not establish a specific Outlook MCP server or its supported actions. For an Outlook workflow, choose a project that documents Outlook authentication and the mailbox operations you need. A general IMAP-based connection may fit only if the provider and server support the required access; SMTP alone is for sending, not inbox reading.

What happens when a Volanea plan reaches its monthly email limit?

Sending pauses when the project reaches its monthly limit; Volanea does not charge overages. For example, a project on Starter has a 7,500-email monthly allowance, so an application expected to exceed that volume needs a higher-volume plan or a change in its sending plan before the limit is reached.

How much does a free email MCP server cost at higher volumes?

“Free” describes a plan, not a universal cost for every sending level. Volanea includes 1,000 emails each month on its free plan; its listed paid monthly tiers include Starter at $5 for 7,500 emails, Growth at $20 for 50,000, and Scale at $149 for 200,000. MailerSend and Mailtrap publish different allowances and plan structures, so compare their sending and testing limits separately.

Start free with 1,000 emails a month