Mail MCP README

mail-mcp

Production-ready email MCP server for AI agents
IMAP + SMTP + EWS + Microsoft Graph API — built in Rust

Most email MCP servers only do IMAP reads. This one does everything: read, search, send, reply, forward, bulk operations, Microsoft Graph API, and Exchange Web Services — with real OAuth2, multi-account, and multi-provider support. Written in Rust for speed and safety.

What's New in v0.5.0

• New: draft tools (draft_create, draft_update, draft_send, draft_delete, draft_list) closing #3 and #17. An agent can save a message in the account's own Drafts folder, revise it, and send it later, so a person can review it in their mail client first. The same five tools work on every backend: IMAP + SMTP, Exchange Web Services and Microsoft Graph, picked per account (EWS config, else Graph config, else IMAP). list_all_accounts reports the backend as drafts_with.
• Separate gate: MAIL_DRAFTS_ENABLED (default false). It is independent of MAIL_IMAP_WRITE_ENABLED: every operation on an existing draft_id first checks that the target really is a draft (IMAP Drafts folder, EWS IsDraft, Graph isDraft), so enabling drafts never grants move or delete on other mail. draft_send also requires MAIL_SMTP_WRITE_ENABLED.
• Graph accounts need the Mail.ReadWrite scope for drafts; see account setup. The EWS attachment flow (CreateItem with SaveOnly, then CreateAttachment) follows the approach proposed by @verdelyi in #14.
• Fix: SMTP messages now carry a Message-ID on the sender's domain (#7). Before, lettre sent no Message-ID of its own, so the id returned by the send tools was empty and the Sent copy had none.
• Verified live on IMAP (Zoho) and Microsoft Graph; the EWS draft path is covered by unit tests.

What's New in v0.4.17

• New: imap_search_all_mailboxes by @tdabasinskas in #35. Runs one search in every selectable mailbox of an account and returns the newest matches, each labelled with its mailbox. Built for providers without an "All Mail" folder (iCloud), where one conversation is spread over INBOX, Sent Messages and project folders. The work is spread over MAIL_IMAP_SEARCH_CONCURRENCY connections (default 4, clamped to 1..=8); a mailbox that fails is reported and does not abort the search. Scope it with include_mailboxes / exclude_mailboxes.
• Thread search by Message-ID by @tdabasinskas in #34. imap_search_messages takes a new thread_message_id criterion that matches a message and every message whose In-Reply-To or References carries it, so an agent can pull a whole conversation in one search.

What's New in v0.4.16

• imap_list_mailboxes cap is now configurable and visible by @tdabasinskas in #33. The tool used to silently drop everything past 200 mailboxes — on accounts with more folders (e.g. 217 on iCloud), folders like Sent Messages simply never appeared. The cap is now set with MAIL_IMAP_MAX_MAILBOXES (default 200, clamped to 1..=10000), and the response carries two additive fields, total and truncated, so clients can always tell a capped list from a complete one. Nothing changes unless the variable is set.

What's New in v0.4.15

• New: opt-in streamable HTTP transport by @tdabasinskas in #32. Set MAIL_MCP_TRANSPORT=http to serve MCP streamable HTTP (stateless) instead of stdio — useful for running the server on another machine behind an MCP gateway. Binds to 127.0.0.1:8000 at /mcp by default, configurable via MAIL_MCP_HTTP_HOST / MAIL_MCP_HTTP_PORT / MAIL_MCP_HTTP_PATH, with graceful shutdown on SIGINT/SIGTERM. The endpoint has no built-in authentication — keep it on loopback or behind an authenticating gateway; see docs/advanced-configuration.md#remote-http-transport. stdio remains the default: nothing changes unless the variable is set.

What's New in v0.4.14

Community release — both changes came from external contributors. Thank you!

• New: MAIL_ATTACHMENT_UPLOAD_DIR — confine outbound file_path attachments by @mjones-PL in #31, closing #11. Without a restriction, send tools would read any file the server process can access — a prompt-injected model could exfiltrate SSH keys or .env files as attachments. Set this variable to a directory and every file_path is canonicalized (.. and symlinks resolved) and must land inside it; missing and out-of-scope files return the same error so the check cannot probe for file existence. Opt-in: when unset, behavior is unchanged. See docs/security.md#outbound-attachment-scope.
• Fixed: FROM_EMAIL validation no longer rejects dotless internal hosts by @arwack in #30. Addresses the v0.4.13 upgrade caveat: noreply@localhost / alerts@intranet style addresses on corporate internal relays are accepted again, while the real typo checks (multiple @, whitespace, empty local part) remain. If you held the v0.4.13 upgrade because of this, v0.4.14 is safe.

What's New in v0.4.13

• effective_from() helper + FROM_EMAIL startup validation by @arwack in #29 — the follow-up to their #19. The from_email → user fallback now lives in one place (SmtpAccountConfig::effective_from()), and MAIL_SMTP_<ID>_FROM_EMAIL is validated when the server starts instead of failing on the first send.
• Behavior change — read before upgrading: a malformed FROM_EMAIL (multiple @, whitespace, empty local part, or a domain without a dot) now prevents the server from starting, for all accounts. Note that dotless domains such as user@localhost or alerts@intranet are currently rejected too; if you use an internal relay address like that, hold the upgrade — a follow-up relaxing the dot rule is under discussion in #29.
• Hardened APPEND wire-format tests by @tordable in #28: mailbox quoting, announced literal length, and byte-for-byte payload are now asserted for every append test.

What's New in v0.4.12

Community release — both changes came from external contributors. Thank you!

• iCloud mailbox aliases + reads no longer mark messages as read by @felipefdl in #16. Short mailbox names now resolve to each provider's real folder (Sent → Sent Messages on iCloud / [Gmail]/Sent Mail on Gmail, Trash → Deleted Messages, and so on, multi-language) in search, copy, and move. Raw message fetches now use BODY.PEEK[], so reading a message through the MCP no longer sets \Seen as a side effect — with a BODY[] fallback for servers that reject PEEK (the deprecated RFC822 item, removed in #23, stays out). Validated against a real iCloud mailbox by the author; includes alias-resolution tests and iCloud setup docs.
• Optimized multi-stage Dockerfile + docker-compose by @monssefbaakka in #5. cargo-chef layer caching, TARGETARCH-aware musl cross-builds (amd64/arm64), and a scratch runtime image — 16.9 MB, down from 25.5 MB — verified to respond to MCP initialize/tools-list over stdio. The toolchain pin was bumped to Rust 1.90 (the codebase's let-chains require >= 1.88).

What's New in v0.4.11

Community bugfix release — both fixes came from external contributors. Thank you!

• Fixed: save-to-Sent silently failed on strict IMAP servers (iCloud and others) by @dominikknafelj in #26, reported in #25. The \Seen flag introduced in v0.4.10 was sent without the RFC 3501 parenthesized flag-list syntax (APPEND "Sent" \Seen … instead of APPEND "Sent" (\Seen) …), because async-imap interpolates the flags argument verbatim. Strict servers rejected the APPEND and the sent copy was lost — while the tool still reported status: ok. Flags are now normalized before hitting the wire, and smtp_send_message / smtp_reply_message / smtp_forward_message responses include a new saved_to_sent field (true/false, or null when saving is disabled) so callers can detect archival failures. @tordable diagnosed and fixed the same root cause concurrently in #24.
• Fixed: message reads returned empty on iCloud by @tdabasinskas in #23. Raw message fetches used the deprecated RFC822 item, which iCloud accepts but leaves unpopulated. Fetches now use the IMAP4rev1 BODY[] item — same \Seen semantics, works everywhere — with a mock-server regression test pinning the wire format.

What's New in v0.4.10

Community release — all three changes came from external contributors. Thank you!

• NetEase IMAP compatibility (126.com / 163.com / yeah.net) by @pep-27 in #21. NetEase servers reject mailbox access from clients that don't identify themselves. mail-mcp now sends the RFC 2971 ID command after authentication whenever the server advertises the ID capability. Includes mock-server regression tests and NetEase setup docs in docs/account-setup.md.
• MAIL_SMTP_<ID>_FROM_EMAIL — sender address override by @arwack in #19. For shared/group mailboxes where SMTP authenticates with a personal account but the From address should be the group address. Applies to send, reply (including reply-all self-address detection) and forward; falls back to _USER when unset.
• Sent-mail copies are now marked \Seen by @ray-of-darkness in #9. Copies the MCP appends to the Sent folder after SMTP send no longer show up as unread.

What's New in v0.4.9

• New tool imap_get_attachment — download a single attachment to disk. Until now the only ways to reach attachment bytes were imap_get_message (which returns attachment metadata and optional extracted PDF text, never the binary) and imap_get_message_raw (capped at 1 MB and base64-encoded into the response). A 7 MB email with X-ray images could not be retrieved at all — over the cap, and dumping it into the response would blow up the model's context anyway.
• How it works: call imap_get_attachment with the message_id plus a selector — either part_id (the value imap_get_message reports for each attachment) or filename. The server fetches the full message (no size cap on the server side), extracts and decodes just that one part, and writes it to disk, returning { file_path, filename, content_type, part_id, size_bytes }. The binary never enters the response, so context stays small. The saved path feeds straight into a local reader (e.g. an image-description tool or a PDF reader).
• Where files land: output_dir argument if given, else the MAIL_ATTACHMENT_DOWNLOAD_DIR environment variable, else the system temp dir. Filenames are sanitized (basename only, control characters stripped) to prevent path traversal, and prefixed with the message UID and part id to avoid collisions.
• Optional inline base64: set include_base64: true to also get the bytes in the response, but only when the attachment is at most max_inline_bytes (default 256 KiB). Off by default.

What's New in v0.4.8

• SAVE_SENT is now per-account with a provider-aware default. Previously, saving a copy of outgoing mail to the Sent folder via IMAP APPEND was controlled by a single global flag, MAIL_SMTP_SAVE_SENT. The problem: providers that already save sent mail server-side (Gmail, Zoho) ended up with two identical copies in Sent, while a generic SMTP server or Office 365 (which do not auto-save on SMTP submission) lost the copy entirely when the flag was false.
• Provider-aware default (when nothing is configured):
  • Gmail (smtp.gmail.com): saves server-side and deduplicates by Message-ID → the MCP does not append (false).
  • Zoho (smtp.zoho.com): saves server-side but does not deduplicate → the MCP does not append (false), avoiding the duplicate.
  • Office 365 / generic SMTP: do not auto-save on SMTP submission → the MCP does append (true), or the sent copy would be lost.
• Per-account override: MAIL_SMTP_<ID>_SAVE_SENT=true|false takes priority over everything. The global MAIL_SMTP_SAVE_SENT still works as a coarse override (wins over the provider default, loses to the per-account override).
• Precedence: per-account → global → provider-aware default.

What's New in v0.4.7

• Critical fix — graph_send_message silently dropped attachments on threaded replies. When called with in_reply_to + attachments, the createReply → PATCH → send flow included the attachments in the PATCH against /me/messages/{id}. Microsoft Graph treats Message.attachments as a navigation property and silently discards the field on PATCH (2xx response, no error), so the message went out as single-part text/html with no file. The MCP returned status: ok and the caller assumed success. Invisible data loss.
• The fix: in send_via_reply(), attachments are now uploaded one by one to POST /me/messages/{draft_id}/attachments between the PATCH and the send. Files < 3 MB go inline (JSON with base64 contentBytes); files ≥ 3 MB use createUploadSession with 4 MB chunked PUTs. The attachments field was removed from the PatchDraftRequest struct so the regression cannot be reintroduced by a type-correct edit.
• No change to flows that already worked. send_via_sendmail (new messages without in_reply_to) uses POST /me/sendMail with attachments inline in the JSON — Graph DOES accept the field on that endpoint and never dropped it. That path is untouched.
• Regression test added: patch_draft_request_never_serializes_attachments fails if anyone re-adds the field to the struct.
• Reference: BUG_GRAPH_ATTACHMENTS.md at the repo root documents the full reproduction, root cause, and the empirical evidence behind the fix.

What's New in v0.4.6

• Server-side enforcement of HARD RULE #1. Three releases of prompt-only hardening (v0.4.3 → v0.4.4 → v0.4.5) still left LLMs occasionally leaking literal </body_text><parameter name="body_html"> markup into the recipient's inbox. v0.4.6 adds a real validator that rejects the tool call before any SMTP / Graph / EWS attempt if body_text or body_html contains tool-call wrapper syntax. The check is wired into all 5 send paths (smtp_send_message, smtp_reply_message, smtp_forward_message, graph_send_message, ews_send_message).
• The forbidden markers are case-insensitive and tightly scoped — only the pseudo-tags that have no legitimate use in human correspondence: <body_text>, </body_text>, <body_html>, </body_html>, <function_calls>, </function_calls>, <invoke name=, </invoke>, and <parameter name="body_*">. Generic technical content that happens to mention <parameter> for an XML schema or <invoke> in a code example still passes.
• HARD RULE #1 wording updated to announce the server-side rejection, so the LLM knows it's a hard contract — not a suggestion it can ignore.
• No breaking changes for clean callers: well-behaved messages send exactly as before.

What's New in v0.4.5

• serverInfo now reports name="mail-mcp" + the crate version (the framework previously returned its own rmcp 0.16.0, which never changes between releases). Useful for verifying the active version with /mcp, and so any client-side cache keyed by (server, version) invalidates on each bump.
• MCP instructions reorganized: the 3 critical anti-concatenation rules (which in v0.4.3 and v0.4.4 sat at the end of the block and could be lost to truncation / diluted attention) now appear as HARD RULE #1, #2, #3 at the TOP, right after the title. Consolidated into 3 short paragraphs (previously 3 long sections, ~1500 characters combined).
• No functional changes to the server. Same SMTP/IMAP/EWS/Graph, same tool set, same behavior. Only the text exposed to the client changed.

Important for these rules to take effect

Clients that resume a session with claude --continue (or /resume) do NOT refresh the MCP system_prompt — they keep the one from that session's first handshake. If your session predates v0.4.5, the rules won't reach your context even if the on-disk binary is updated. To receive them, start a NEW session in the project (not --continue).

What's New in v0.4.4

• Preview hygiene rule in MCP instructions: when the LLM shows the user the email preview before sending, it should render ONE clean version of the body (markdown-style bullets, bold, links as text + URL) and state that the message will go multipart — but it must NOT dump the raw HTML source (<p>, <strong>, <a href>...) into the preview. Two reasons:

  1. The human reviewer wants to read the message, not audit markup — showing the HTML is noise.
  2. Exhibiting both the plain-text string AND the HTML string side by side in the preview is exactly the context that has historically led LLMs to concatenate them in the eventual tool call (the bug v0.4.3 documented). Hiding the HTML source from the preview removes the temptation.

  Complements the PREVIEW DOES NOT EQUAL TOOL CALL rule introduced in v0.4.3.

What's New in v0.4.3

• Server-side guidance against malformed tool calls. The MCP instructions block now explicitly tells the calling LLM that body_text and body_html are TWO SEPARATE JSON fields and must NEVER be concatenated. Previous wording ("send BOTH body_text AND body_html") was ambiguous and some LLMs interpreted it as "concatenate both with <body_text>...</body_html> pseudo-tags inside a single body_text string". When that happens, the recipient sees garbled duplicated content, AND any later Claude session that opens the saved copy via this MCP gets a Usage Policy block (the leaked <invoke>...</invoke> looks like a prompt-injection attempt to safety filters). The new instruction shows a CORRECT vs WRONG example and bans pseudo-tags / tool-call wrapper syntax inside email fields.

What's New in v0.4.2

• Release pipeline fixed: the publish-npm job in the CI release workflow has been disabled. It was inherited from the upstream fork and tried to publish to @bradsjm/mail-imap-mcp-rs, a scope this org does not own — every release was 404-ing on that step. See "Releasing" below for the full explanation and how to re-enable npm publishing if needed.
• Auto-trigger releases on tag push: .github/workflows/release.yml now fires on push: tags: ['v*'], so tagging vX.Y.Z and pushing is all it takes to cut a release. workflow_dispatch is retained as a manual escape hatch.
• Cleanup: removed the dangling init-npm-placeholder.yml workflow (also referenced the fork's npm scope).
• docs: README gains a "Releasing" section documenting the new flow and the npm decision.

What's New in v0.4.1

• Fix: save_to_sent_folder now archives the exact RFC822 bytes that were sent (via lettre.formatted()), instead of a hand-rolled text-only stub. The Sent-folder copy keeps the HTML body, the multipart/alternative structure, and the RFC 2047-encoded subject — no more ??? where accents used to be, and HTML is no longer silently dropped.
• Improved: localized Sent-folder detection — Enviado[s], Elementos enviados, Enviadas, Itens enviados, Envoyés, Éléments envoyés, Gesendet, Posta inviata, Verzonden, Wysłane, plus nested variants. Previously only English names were recognized, so Zoho/localized IMAP accounts fell through to a non-existent "Sent" folder.
• Improved: smtp_forward_message accepts body_html (was hardcoded to plain-text only).
• Improved: EWS send gains bcc, in_reply_to, references (via <t:InternetMessageHeaders>), plus full recipient + subject-length validation — now at parity with the SMTP and Graph send paths.
• Improved: Graph API threading fallbacks now log. WARN when the message-lookup HTTP call fails (rate limit, 5xx, permissions) so operators see threading degraded due to a real error; DEBUG when the original message is legitimately not found.
• Refactor: EWS XML parsing migrated from substring matching to quick-xml. Fixes a latent namespace-collision bug (<soap:Body> vs <t:Body>), correctly decodes XML entities and CDATA, and handles attribute values containing = (common in base64-like EWS item IDs).
• Cleanup: zero warnings on cargo build --release.
• Tests: 64 (up from 47).

Why This Project

Feature Matrix

EWS is the simplest way to add Microsoft accounts — single OAuth2 token for both reading and sending. Works even on tenants that block Graph API and IMAP.

Quickstart — Let Claude Code do it

Copy and paste this prompt into Claude Code and it will install, compile, and configure everything for you:

Install and configure the mail-mcp MCP server from https://github.com/tecnologicachile/mail-mcp

1. Clone the repo, build with cargo build --release
2. Add the MCP server to .claude.json with the binary path
3. For Microsoft accounts: use EWS (simplest) — run device code flow with
   client_id d3590ed6-52b3-4102-aeff-aad2292ab01c and scope
   https://outlook.office365.com/EWS.AccessAsUser.All offline_access
   Then configure MAIL_EWS_<ID>_USER and MAIL_EWS_<ID>_REFRESH_TOKEN
4. For Gmail: configure MAIL_IMAP + MAIL_SMTP with App Password from
   https://myaccount.google.com/apppasswords
5. For Zoho: configure MAIL_IMAP + MAIL_SMTP with standard password
6. Enable write/send: MAIL_IMAP_WRITE_ENABLED=true, MAIL_SMTP_WRITE_ENABLED=true

My email accounts to configure:
- <your-email@example.com>

Replace the last line with your email(s). Claude Code will guide you through each step including the OAuth2 device code flow for Microsoft accounts.

Manual Setup (2 minutes)

Install a prebuilt binary:

# macOS and Linux (Homebrew)
brew install tecnologicachile/tap/mail-mcp

# macOS and Linux (shell installer)
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/tecnologicachile/mail-mcp/releases/latest/download/mail-mcp-installer.sh | sh

# Windows
powershell -ExecutionPolicy Bypass -c "irm https://github.com/tecnologicachile/mail-mcp/releases/latest/download/mail-mcp-installer.ps1 | iex"

Or build from source:

git clone https://github.com/tecnologicachile/mail-mcp.git
cd mail-mcp
cargo build --release

Add to your MCP client config (Claude Code, Cursor, etc.). With a prebuilt binary, use "command": "mail-mcp":

{
  "mcpServers": {
    "mail": {
      "command": "./target/release/mail-mcp",
      "env": {
        "MAIL_IMAP_DEFAULT_HOST": "imap.gmail.com",
        "MAIL_IMAP_DEFAULT_USER": "you@gmail.com",
        "MAIL_IMAP_DEFAULT_PASS": "your-app-password",
        "MAIL_SMTP_DEFAULT_HOST": "smtp.gmail.com",
        "MAIL_SMTP_DEFAULT_PORT": "587",
        "MAIL_SMTP_DEFAULT_USER": "you@gmail.com",
        "MAIL_SMTP_DEFAULT_PASS": "your-app-password",
        "MAIL_SMTP_DEFAULT_SECURE": "starttls",
        "MAIL_IMAP_WRITE_ENABLED": "true",
        "MAIL_SMTP_WRITE_ENABLED": "true"
      }
    }
  }
}

That's it. Your AI agent can now read, search, send, reply, and manage emails.

Microsoft Account? Use Graph API

Microsoft blocks SMTP on personal accounts. Use Graph API instead:

{
  "env": {
    "MAIL_IMAP_DEFAULT_HOST": "outlook.office365.com",
    "MAIL_IMAP_DEFAULT_USER": "you@hotmail.com",
    "MAIL_IMAP_DEFAULT_PASS": "your-app-password",
    "MAIL_OAUTH2_DEFAULT_PROVIDER": "microsoft",
    "MAIL_OAUTH2_DEFAULT_CLIENT_ID": "9e5f94bc-e8a4-4e73-b8be-63364c29d753",
    "MAIL_OAUTH2_DEFAULT_CLIENT_SECRET": "none",
    "MAIL_OAUTH2_DEFAULT_REFRESH_TOKEN": "<your-token>"
  }
}

Get your token in 1 minute with device code flow. See Account Setup Guide.

37 MCP Tools

Read (10 tools)

Write (11 tools)

Send (5 tools)

EWS — Exchange Web Services (3 tools)

Drafts (5 tools)

The agent prepares the email, a person reviews it in their mail client (or asks the agent to send it). One set of tools for every account: drafts go to the IMAP Drafts folder, or through EWS or Microsoft Graph when the account is configured for them (list_all_accounts shows drafts_with).

Enable with MAIL_DRAFTS_ENABLED=true. It is independent of MAIL_IMAP_WRITE_ENABLED: with only drafts enabled, the agent can write drafts but cannot move, flag or delete other mail, and every draft tool refuses ids that are not drafts. draft_send additionally needs MAIL_SMTP_WRITE_ENABLED=true. Graph accounts need the Mail.ReadWrite scope.

Attachments

Send files with any send tool. Two modes:

// Large files — MCP reads from disk (recommended)
"attachments": [{"file_path": "/path/to/report.pdf"}]

// Small files — inline base64
"attachments": [{"filename": "note.txt", "content_type": "text/plain", "content_base64": "SGVsbG8="}]

Filename and MIME type are auto-detected from the file path. Reply with include_original_attachments: true to forward original attachments.

Downloading an attachment from a received message: use imap_get_attachment with the message_id and a part_id (from imap_get_message) or filename. It writes the decoded file to disk and returns the path — no size cap, and the binary stays out of the response. Set the default download directory with MAIL_ATTACHMENT_DOWNLOAD_DIR (falls back to the system temp dir), or pass output_dir per call.

To limit which local files send tools may attach via file_path, set MAIL_ATTACHMENT_UPLOAD_DIR; paths outside it (including via .. or symlinks) are rejected. See docs/security.md.

Bulk Operations (2 tools)

Setup Helper (1 tool)

Multi-Account

Configure as many accounts as you need:

# Gmail
MAIL_IMAP_GMAIL_HOST=imap.gmail.com
MAIL_IMAP_GMAIL_USER=me@gmail.com
MAIL_IMAP_GMAIL_PASS=app-password

# Apple iCloud (App-Specific Password from appleid.apple.com)
MAIL_IMAP_ICLOUD_HOST=imap.mail.me.com
MAIL_IMAP_ICLOUD_USER=you@icloud.com
MAIL_IMAP_ICLOUD_PASS=app-specific-password
MAIL_SMTP_ICLOUD_HOST=smtp.mail.me.com
MAIL_SMTP_ICLOUD_USER=you@icloud.com
MAIL_SMTP_ICLOUD_PASS=app-specific-password
MAIL_SMTP_ICLOUD_SECURE=starttls

# Microsoft 365
MAIL_IMAP_WORK_HOST=outlook.office365.com
MAIL_IMAP_WORK_USER=me@company.com
MAIL_OAUTH2_WORK_PROVIDER=microsoft
MAIL_OAUTH2_WORK_CLIENT_ID=your-client-id
MAIL_OAUTH2_WORK_CLIENT_SECRET=none
MAIL_OAUTH2_WORK_REFRESH_TOKEN=your-token

# Zoho
MAIL_IMAP_DEFAULT_HOST=imap.zoho.com
MAIL_IMAP_DEFAULT_USER=info@mydomain.com
MAIL_IMAP_DEFAULT_PASS=password
MAIL_SMTP_DEFAULT_HOST=smtp.zoho.com
MAIL_SMTP_DEFAULT_USER=info@mydomain.com
MAIL_SMTP_DEFAULT_PASS=password
MAIL_SMTP_DEFAULT_SECURE=starttls

Use account_id in tool calls: "account_id": "gmail", "account_id": "icloud", "account_id": "work", "account_id": "default".

Security

• TLS enforced on all connections (except localhost proxies)
• Passwords in SecretString — never logged or returned in responses
• Write operations gated — require explicit MAIL_IMAP_WRITE_ENABLED=true
• Send operations gated — require explicit MAIL_SMTP_WRITE_ENABLED=true
• Drafts gated separately — MAIL_DRAFTS_ENABLED=true allows writing drafts without any other write access
• Delete confirmation — requires confirm: true
• HTML sanitized with ammonia (prevents XSS)
• Bounded outputs — body text, HTML, attachments truncated to configurable limits
• OAuth2 tokens cached with 10-minute refresh margin
• No secrets in responses — credentials never exposed via MCP tools
• HTTP transport is opt-in and unauthenticated — stdio by default; MAIL_MCP_TRANSPORT=http binds to loopback unless told otherwise, and belongs behind an authenticating gateway

Configuration Reference

Roadmap

• IMAP read operations (search, fetch, parse)
• IMAP write operations (copy, move, delete, flags)
• IMAP bulk operations (up to 500 per call)
• Cursor-based pagination with TTL
• SMTP send, reply, forward
• Microsoft Graph API (sendMail)
• OAuth2 XOAUTH2 (Google + Microsoft)
• Separate Graph API tokens for enterprise
• Multi-account via environment variables
• PDF text extraction from attachments
• HTML sanitization (ammonia)
• Provider setup documentation with direct links
• Attachment sending (SMTP/Graph)
• Reply with original attachments
• CDATA sanitization (Zoho bug fix)
• Email confirmation protocol (preview before send)
• Token-optimized instructions (75% reduction)
• On-demand setup guide tool
• EWS (Exchange Web Services) — single token for read + send on Microsoft
• EWS with Microsoft Office Client ID (works on restricted tenants)
• Graph API threading — createReply flow for proper conversation threading
• HTML formatting guidance — LLM prefers multipart (text + HTML) for human emails
• — byte-identical copy of what the recipient received (v0.4.1)

Next — Local cache with instant search

• SQLite + FTS5 local email cache — instant searches (<10ms vs 3-10s)
• Incremental sync — UIDVALIDITY + last UID delta sync
• Connection pooling — persistent IMAP sessions per account
• Cross-account search — search all accounts at once
• Email statistics — counts, top senders, activity by date

Future

• Docker image
• npm/npx distribution
• Draft management
• Contact search
• IMAP IDLE (real-time notifications)
• Hosted documentation site

Documentation

Development

cargo test              # 64 unit + integration tests
cargo fmt -- --check    # formatting
cargo clippy --all-targets -- -D warnings  # linting

See AGENTS.md for contributor guidelines.

Releasing

Releases are automated via cargo-dist. To ship a new version:

1. Bump version = "X.Y.Z" in Cargo.toml (the release workflow enforces that this matches the pushed tag).
2. Commit the bump + any release notes to main.
3. Tag and push:
   bashCopy

   git tag vX.Y.Z
   git push origin main --tags
   

4. The push: tags: ['v*'] trigger in .github/workflows/release.yml compiles binaries for Linux / macOS (Intel + Apple Silicon) / Windows, generates installer scripts (.sh, .ps1), creates the GitHub Release, and attaches all artifacts with SHA256 checksums.
5. If anything fails you can re-run the workflow manually from the Actions tab (the workflow_dispatch trigger is preserved as an escape hatch).

npm publishing is intentionally disabled. The upstream fork was configured to publish as @bradsjm/mail-imap-mcp-rs, a scope this organization does not own, which caused every release to 404 on npm publish. The npm tarball is still generated and attached to each GitHub Release so users can install via npm install ./mail-mcp-npm-package.tar.gz manually. To enable npm registry publishing for this fork: create an npm org (e.g. @tecnologicachile), configure Trusted Publishing on npmjs.com pointing at this repo, set publish-jobs = ["npm"] in dist-workspace.toml, and run dist generate --allow-dirty to restore the publish-npm job in release.yml.

Contributing

Contributions welcome! Check out the issues for good first issues.

If mail-mcp is useful to you, a ⭐ on the repo helps others discover it.

License

MIT License — see LICENSE for details.