Token导航 LogoToken导航TokenDH.com
研究检索敏感数据clawhub未标认证来源可访问clear审计通过

mailbird-mcpmailbird MCP 搜索

Agent Skill

mailbird-mcp 用于查找、检索和筛选相关信息,适合在 OpenClaw 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

1,929

周安装

82

GitHub Stars

公开资料未说明

下载量

676
OpenClaw

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

MIT-0

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:mailbird-mcp(mailbird MCP 搜索)
来源仓库:https://github.com/mailbirdagent/mailbird-mcp
安装命令:
openclaw skills install mailbird-mcp
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 OpenClaw 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

ClawHubOpenClaw
openclaw skills install mailbird-mcp

简介

调用本地 Mailbird 客户端的 MCP 服务执行各类邮件操作。

  • 适合已使用 Mailbird 作为主要邮箱工具的用户进行深度集成。
  • 支持收件箱管理、草稿编辑、附件下载和邮件搜索等功能。
  • 依赖本地 Mailbird 运行且开启 MCP 接口,需确保软件版本兼容。
  • mailbird-mcp 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

SKILL.md

name
mailbird-mcp
description
Use the Mailbird MCP server (running locally inside the Mailbird email client) for any email-related task — inbox triage, sending, search, drafts, attachments, contacts. Never grep code for mail content.
homepage
https://www.getmailbird.com
metadata
{"openclaw":{"emoji":"📬","requires":{"env":["MAILBIRD_MCP_URL","MAILBIRD_MCP_TOKEN"]},"primaryEnv":"MAILBIRD_MCP_TOKEN","envDescriptions":{"MAILBIRD_MCP_URL":"Optional. Local Mailbird MCP endpoint (default: http://127.0.0.1:18790/mcp). Must point at 127.0.0.1 / localhost only.","MAILBIRD_MCP_TOKEN":"Bearer token from Mailbird → Settings → Wingman AI → Copy token. Treat as a credential to your mailbox; never share with remote agents."},"mcp":{"name":"mailbird","transport":"http","urlEnv":"MAILBIRD_MCP_URL","urlDefault":"http://127.0.0.1:18790/mcp","tokenEnv":"MAILBIRD_MCP_TOKEN","headerName":"Authorization","headerPrefix":"Bearer "}}}

Mailbird MCP

You have access to a local Mailbird MCP server that exposes the user's real email accounts, folders, conversations, drafts, and attachments. It's running inside the user's Mailbird desktop app on 127.0.0.1 only — there is no remote variant.

The single most important rule: for ANY task that involves email, inbox, messages, drafts, contacts, attachments, folders, or sending — even when the user phrases it casually ("check my inbox", "any reply from X yet?", "draft a reply to that invoice", "find that thread from Mira") — reach for these tools first. Do not grep the local filesystem, do not read code, do not guess. The server is the source of truth.

Setup

The user enables the MCP server inside Mailbird at: Settings → Wingman AI → Enable MCP server.

That tab also exposes:

  • The bearer token (Copy button).
  • The endpoint URL (http://127.0.0.1:<port>/mcp, port shown next to status).
  • The "Allow write actions" toggle — required before any write tool will work.

If the connection fails, ask the user to verify the toggle is on and that they've copied the current token. Tokens regenerate when the server is disabled and re-enabled.

Configuration (environment variables)

Both are optional; defaults work for a single-user local install.

VariableRequiredDefaultNotes
MAILBIRD_MCP_URLoptionalhttp://127.0.0.1:18790/mcpLocal Mailbird MCP endpoint. Must be 127.0.0.1 / localhost only.
MAILBIRD_MCP_TOKENoptionalBearer token from Mailbird's Wingman AI tab. If unset and Mailbird's settings file is reachable, the agent reads it from there; otherwise the agent will prompt.

Security model

This skill grants the agent access to the user's full mailbox: message bodies, attachments, contacts, and the ability to send mail (when the write-action gate is on). Treat the URL and token accordingly:

  • The Mailbird MCP server only binds to loopback (127.0.0.1).

Don't proxy, port-forward, or tunnel it to a public address. Don't paste the URL or token into any remote / cloud-hosted agent that doesn't run on the same machine as Mailbird.

  • The token is a credential equivalent to mailbox login. Don't echo it

into chat transcripts, commit it, share it in screenshots, or include it in bug reports. Tokens regenerate when the server is disabled and re-enabled — rotate immediately if it leaks.

  • Write actions (archive, trash, send, etc.) require the user to flip

Allow write actions in the Wingman AI tab. Sending additionally requires per-call confirm: true. The skill should always show drafts to the user before sending.

  • Mailbird's optional Audit log of MCP requests records every call

(method + params, never responses) to a local file the user can inspect. Recommend they enable it for visibility.

Start-of-session checklist

Run these the first time you touch the server in a session, before any non-trivial action:

  1. Read mailbird://help via resources/read. It's the canonical

user guide — covers the ID model, write-tool gating, send pipeline, search index lag, archive→restore, attachment handling, inline images. Skim it once and remember the key recipes.

  1. list_accounts to learn the configured account ids.
  2. For folder-scoped work, list_folders(accountId) and pick by the

identity field — Inbox, Sent, Drafts, Trash, Spam, Archived, AllMail, Generic (user-created). Folder ids are NOT stable across accounts. list_accounts does not return the inbox folder id — always discover via list_folders.

Read tools (always available)

  • list_accounts — accounts with id, sender name, email, unread count.
  • list_folders(accountId) — folders for one account.
  • list_conversations(folderId, limit?, unreadOnly?, starredOnly?, importantOnly?) — recent conversations in a folder.
  • get_conversation(conversationId, folderId) — message list + metadata for one thread.
  • get_message(messageId) — full message body, with cid: images rewritten to mailbird://messages/{messageId}/attachments/{attachmentId} resource URIs.
  • get_unread_counts(accountId? | folderId?) — quick triage signal.
  • search_conversations(query, accountId?, folderId?) — Mailbird search syntax (from:foo subject:bar). Results carry actualFolders[]; use those ids to act on hits, not the virtual folderId: -2.
  • list_attachments(messageId) / get_attachment_status(...) / get_attachment_content(...).
  • get_send_status(messageId)sent / draft_pending_send / scheduled / trashed.

Write tools (gated by "Allow write actions")

  • archive_conversation, trash_conversation, move_conversation, move_conversation_to_inbox.
  • mark_conversation_as_read / unread, flag_conversation_important, star_conversation / unstar_conversation, mark_conversation_as_spam / unmark_conversation_as_spam, snooze_conversation(wakeAtUtc).
  • create_draft(accountId, to, cc?, bcc?, subject, body, attachments?) — saves a draft, returns messageId. Does NOT send.
  • update_draft(messageId, ...) — replace any field on an existing draft.
  • reply_to_conversation, reply_all_to_conversation, forward_conversation — create a draft with the standard quoted scaffold and return messageId. Do NOT send. Body is up to you to finalise.
  • send_message_now(messageId | accountId+to+...; confirm: true) — actually sends. Always show the draft to the user and get explicit approval first. Returns status: "queued" plus a deliveryState field signalling IMAP/SMTP health.
  • unsubscribe_from_newsletter(messageId) — uses the List-Unsubscribe header. Returns structured "not_applicable" / "already_unsubscribed" when relevant.
  • delete_conversation_permanentlyonly applies to conversations currently in Trash or Spam. From elsewhere, trash first then re-discover the new id and call this on the trash copy.

If a write tool returns an error pointing at the "Allow write actions" toggle, surface it to the user verbatim — do not retry.

Pitfalls (these bite less-careful agents)

  1. Conversation IDs are per-folder. After trash_conversation,

archive_conversation, or move_conversation, the conversation has a NEW id in its destination folder. Re-discover via list_conversations(folderId=<destination>) before chaining further actions. Message ids, on the other hand, are stable across folders.

  1. Search index lag (~10–30s). A message you just sent or received may

not be in search_conversations results yet. For very recent items, prefer list_conversations(folderId=<sent_folder>) over searching.

  1. Send pipeline. send_message_now returns immediately with

status: "queued". The message stays briefly visible in Drafts before moving to Sent — that's normal. Use get_send_status to confirm.

  1. Archive destination depends on the provider. Gmail and IMAP-with-labels

accounts archive into the AllMail folder; everything else uses Archived. Exactly one will exist per account. The full restore recipe (list_folders → list_conversations → move_conversation_to_inbox) lives in mailbird://help.

  1. Inline (cid:) images in get_message results are rewritten to

mailbird://messages/.../attachments/... URIs. Resolve via resources/read. The response also carries an inlineAttachments map.

Reply pattern

Standard chain for an agent-authored reply:

1. reply_to_conversation(conversationId, folderId)        → messageId
2. update_draft(messageId, body: "<your prose>")           # quoted scaffold preserved
3. <show draft to user, get approval>
4. send_message_now(messageId, confirm: true)              → status: queued
5. (optional, ~5s later) get_send_status(messageId)        → status: sent

For a brand-new message (no thread), use create_draft with to/subject/body/attachments directly, then steps 3–5.

When to escalate to the user

  • Any write before "Allow write actions" is enabled.
  • Any send — always show the draft and get approval.
  • Permanent delete from anywhere other than Trash/Spam.
  • Search returns nothing for content the user expects to exist (could be

index lag, suggest the user wait and retry).

When uncertain about provider-specific behaviour or an edge case, read mailbird://help again — it's the authoritative source.

适合场景

01

OpenClaw 用户查找和安装 Skill 时

02

用户想查找某类 Agent Skill 时

03

需要根据任务场景推荐可安装能力包时

04

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

补充不同宿主或平台的使用分布数据

能力 5

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

OpenClaw

87.1%
按下载量换算589

安全审计

VirusTotal

通过

ClawScan

通过

Static analysis

通过

权限和风险

敏感数据

该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills