Help Scout MCP Server

io.github.drewburchfield由 GitHub 账号 drewburchfield 发布v1.7.0更新于 Oct 24, 2026

Search Help Scout conversations, customers, organizations, threads, and inboxes with AI assistants

已验证STDIO仅桌面知识与记忆商业与电商

概览

AI 生成的概览

本地运行的 MCP 服务器,让 AI 助手搜索和读取 Help Scout 的收件箱、会话、客户、组织、会话线程和 Docs 文章。

可以做什么
按关键词、日期范围、状态、标签、邮箱域名或工单号搜索和列出会话,并跨多个收件箱监控动态。按姓名或邮箱查找客户,查看客户资料和联系方式。浏览组织,并遍历其成员和会话历史。读取会话详情、摘要、完整线程历史、附件和收件箱路由。搜索并获取 Help Scout Docs 知识库文章。在运营者开启后,可选地创建回复草稿、内部备注、标签、状态、指派和稍后提醒变更,面向客户的发送需再开启第二个开关。
适用场景
适合在 Help Scout 中处理客户会话的支持团队,希望助手在编辑器或聊天窗口中起草回复时调取工单上下文、客户与账户历史或知识库文章。默认安装为只读,适合在考虑写入权限之前进行查询和分析。
运行要求
通过 npx(Node.js)或 Docker 本地运行,也可作为 Claude Desktop 扩展安装。需要 Help Scout 私有应用:HELPSCOUT_APP_ID 和 HELPSCOUT_APP_SECRET(也接受 HELPSCOUT_CLIENT_ID 和 HELPSCOUT_CLIENT_SECRET)。Docs 知识库工具可选需要 HELPSCOUT_DOCS_API_KEY。需要访问 Help Scout API 的网络连接。
安装前请注意
应用以创建它的用户身份及其权限进行认证,没有单独的权限范围选择。默认包含消息正文;REDACT_MESSAGE_CONTENT 可隐藏正文,但不是合规边界,也不会移除所有客户标识。写入默认关闭;HELPSCOUT_ENABLE_WRITES 会增加草稿、备注、标签、状态和指派变更,HELPSCOUT_ENABLE_CUSTOMER_VISIBLE_WRITES 会增加 sendReply 和 publishDraft,会向客户发送邮件。删除和管理配置写入不开放。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Help Scout MCP Server,将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。

其他 MCP 客户端

参照 仓库 中的启动说明。

README

[Help Scout MCP Server]

[npm version] [Docker] [Ask DeepWiki] [License: MIT]

An MCP server that gives AI assistants direct access to your Help Scout inboxes, conversations, customers, organizations, threads, and Docs knowledge base. Search tickets, pull customer and account context, inspect articles, spot patterns, and get answers without leaving your editor or chat window.

Built by a Help Scout customer who wanted to give his support team superpowers. If you handle customer conversations in Help Scout and want AI to help you work faster, this is for you.

What You Can Do

  • Search conversations by keyword, date range, status, tag, email domain, or ticket number
  • Look up customers by name, advanced query syntax, or exact email address
  • Explore organizations with direct customer and conversation traversal
  • Inspect conversation detail with raw ticket metadata, summaries, full threads, attachments, and original source
  • Pull full thread history into context before drafting a reply
  • Get conversation summaries with the original customer message and latest staff response
  • Search and retrieve Docs articles from the separate Help Scout Docs API
  • Pull Help Scout reports and metadata for company, conversations, Docs, channels, productivity, happiness, users, teams, system users, statuses, routing, and webhooks
  • Monitor inbox activity across multiple inboxes with a single query
  • Take action with opt-in writes: draft replies, internal notes, tags, status, assignment, snooze, and more, all off by default
  • Reduce message payloads with optional message content redaction and scoped inbox access

Quick Start

Claude Desktop & Claude Cowork (Recommended)

One-click install using Desktop Extensions. One install covers both Chat and Cowork sessions in the Claude desktop app.

  1. Download the latest .mcpb file from releases
  2. Double-click to install (or drag into Claude Desktop)
  3. Enter your Help Scout App ID and App Secret in the extension settings; the settings also carry toggles for message redaction and the opt-in write surface
  4. Restart Claude Desktop

If the tools don't show up in a Cowork session, update the desktop app to the latest version and start a fresh session. (Cowork walkthrough)

Optional: add the helpscout-navigator skill so Claude picks the right operation faster. Go to Customize, click + > Add marketplace from GitHub, enter drewburchfield/help-scout-mcp-server, and install helpscout-navigator.

Claude Code

Register the server, then optionally add the helpscout-navigator skill, which teaches Claude to pick the right operation for each query.

bash
claude mcp add helpscout \  --env HELPSCOUT_APP_ID=your-app-id \  --env HELPSCOUT_APP_SECRET=your-app-secret \  -- npx -y help-scout-mcp-server

Then, for the navigation skill:

  1. Run /plugin marketplace add drewburchfield/help-scout-mcp-server
  2. Run /plugin install helpscout-navigator

The server alone gives you the tools; the skill also teaches the AI how to use them well.

For Cursor, VS Code, and Other MCP Clients

Add to your MCP client's config file (e.g., claude_desktop_config.json, .cursor/mcp.json):

json
{  "mcpServers": {    "helpscout": {      "command": "npx",      "args": ["[email protected]"],      "env": {        "HELPSCOUT_APP_ID": "your-app-id",        "HELPSCOUT_APP_SECRET": "your-app-secret",        "HELPSCOUT_DOCS_API_KEY": "optional-docs-api-key"      }    }  }}

Docker

bash
docker run -e HELPSCOUT_APP_ID="your-app-id" \  -e HELPSCOUT_APP_SECRET="your-app-secret" \  -e HELPSCOUT_DOCS_API_KEY="optional-docs-api-key" \  drewburchfield/help-scout-mcp-server:2.1.0

Getting Your API Credentials

  1. Go to Help Scout > My Apps > Create Private App
  2. Copy your App ID and App Secret

Help Scout uses OAuth2 Client Credentials flow exclusively. Personal Access Tokens are not supported. The app authenticates as the user who created it, with that user's permissions; there is no separate scope selection, which is why the server's own write gating defaults to off.

Help Scout UIEnvironment Variable
App IDHELPSCOUT_APP_ID
App SecretHELPSCOUT_APP_SECRET

Alternative names HELPSCOUT_CLIENT_ID / HELPSCOUT_CLIENT_SECRET are also supported.

Docs knowledge base tools use Help Scout Docs API v1, which is separate from the Mailbox API. Set HELPSCOUT_DOCS_API_KEY only if you want to use listDocs*, searchDocsArticles, getDocsArticle, or redirect tools.

Tools

The server advertises three tools that together reach every supported read operation (55 across the Mailbox and Docs APIs):

ToolPurpose
search_help_scoutFind operations by intent ("customer conversation history", "happiness report")
describe_help_scoutLoad the full input schemas for the operations you selected
read_help_scoutExecute one operation: { "name": "getThreads", "arguments": { ... } }

This keeps the advertised surface small enough that AI clients don't drown in schemas, while every read capability stays one search away. Operations in the current registry also remain callable by name directly. Tool names removed in the v2.0.0 consolidation (for example comprehensiveConversationSearch, structuredConversationFilter, and searchInboxes) are not; their capabilities live in searchConversations and listAllInboxes.

An optional fourth tool, write_help_scout, appears only when an operator turns writes on. See Write operations (opt-in).

For the MCP compatibility contract and roadmap, see:

Which operation should I use?

Run any of these via read_help_scout:

TaskOperationExample
List recent ticketssearchConversations"Show me active tickets from this week"
Find by keywordsearchConversations (contentTerms)"Find conversations about billing errors"
Look up a ticket numbersearchConversations (conversationNumber)"Show me ticket #42839"
Complex filterssearchConversations (emailDomain, tag)"All @acme.com conversations tagged urgent"
Browse customerslistCustomers"Show customers named Jane"
Find a customer by emailsearchCustomersByEmail"Find customer [email protected]"
Inspect a customer profilegetCustomer"Open customer 12345"
Pull customer contact channelsgetCustomerContacts"Show contact details for customer 12345"
Browse organizationslistOrganizations"Show the busiest organizations"
Inspect an organizationgetOrganization"Open organization 456"
List customers in an organizationgetOrganizationMembers"Who belongs to organization 456?"
List organization conversationsgetOrganizationConversations"Show support history for organization 456"
Raw conversation detailgetConversation"Open conversation 12345 with full metadata"
Quick conversation overviewgetConversationSummary"Summarize this conversation"
Full message historygetThreads"Show me the complete thread"
Inspect inbox fields, folders, or routinggetInbox (include)"Show routing for inbox 359402" (include: ["routing"])
Search Docs articlessearchDocsArticles"Find knowledge base articles about refunds"
Retrieve a Docs articlegetDocsArticle"Open Docs article 123"
Current MCP host timegetServerTimeUsed for time-relative searches

Inboxes are auto-discovered when the server connects. AI agents get inbox IDs in their instructions automatically, so no lookup step is needed.

Write operations (opt-in)

A default install is read-only. It advertises the three tools above and nothing else, unchanged from 2.0. Writes exist only after an operator sets a flag.

FlagWhat it adds
HELPSCOUT_ENABLE_WRITES=trueA fourth tool, write_help_scout, carrying 11 tier-1 conversation operations
HELPSCOUT_ENABLE_CUSTOMER_VISIBLE_WRITES=trueTwo more operations on that same tool: sendReply and publishDraft

Tier 1 covers draft replies, internal notes, status changes, assign and unassign, adding and removing tags, custom field values, snooze and unsnooze, and moving a conversation to another inbox. None of it emails anyone: a draft is saved unsent, and a note is visible to teammates only.

Tier 2 is the only path that reaches a customer, and it needs both flags. Every call to sendReply or publishDraft must also carry confirmation naming the operation and the target:

json
{  "name": "sendReply",  "arguments": { "conversationId": "12345", "text": "..." },  "confirm": true,  "confirmOperation": "sendReply",  "targetId": "12345"}

Missing, false, or mismatched confirmation is refused before anything reaches Help Scout. Deletes and admin configuration writes are deliberately not exposed, under any flag.

Set "dryRun": true on any write to validate the arguments and see the exact request that would be sent, without contacting Help Scout.

Full rules: write tool contract.

Configuration

VariableDescriptionDefault
HELPSCOUT_APP_IDApp ID from Help Scout My AppsRequired
HELPSCOUT_APP_SECRETApp Secret from Help Scout My AppsRequired
HELPSCOUT_DEFAULT_INBOX_IDScope searches to a specific inboxNone (all inboxes)
HELPSCOUT_BASE_URLHelp Scout API endpointhttps://api.helpscout.net/v2/
HELPSCOUT_DOCS_API_KEYOptional Docs API key for knowledge base toolsNone
HELPSCOUT_DOCS_BASE_URLHelp Scout Docs API endpointhttps://docsapi.helpscout.net/v1/
REDACT_MESSAGE_CONTENTReplace message bodies with placeholdersfalse
HELPSCOUT_ENABLE_WRITESAdvertise write_help_scout with the tier-1 conversation writesUnset (false)
HELPSCOUT_ENABLE_CUSTOMER_VISIBLE_WRITESAlso enable sendReply and publishDraft, which email the customerUnset (false)
CACHE_TTL_SECONDSCache duration for API responses300
LOG_LEVELLogging verbosity (error, warn, info, debug)info

Compatibility

Works with any MCP-compatible client:

CategoryClients
AI AssistantsClaude Desktop (Chat and Cowork), Goose, and other MCP-enabled assistants
Code EditorsCursor, VS Code, Windsurf, Continue.dev
Command LineClaude Code, Codex, Gemini CLI, OpenCode
CustomAny application implementing the MCP standard

Security and Privacy

Built with security-minded teams in mind:

  • Optional message content redaction. Message bodies are included by default. Set REDACT_MESSAGE_CONTENT=true to replace conversation and thread bodies with placeholders for lower-context analysis. This is not a compliance boundary and does not remove all customer identifiers.
  • Secure authentication. OAuth2 Client Credentials with automatic token refresh.
  • Rate limit handling. Automatic retry with exponential backoff on 429 responses.
  • Scoped access. Optional default inbox configuration limits what the AI can search.

Troubleshooting

Authentication failed? Verify your credentials work with Help Scout directly:

bash
curl -X POST https://api.helpscout.net/v2/oauth2/token \  -d "grant_type=client_credentials&client_id=$HELPSCOUT_APP_ID&client_secret=$HELPSCOUT_APP_SECRET"

Empty search results? Common causes:

  • Forgetting that searchConversations is the single search tool: use contentTerms/subjectTerms for keyword search, plain filters for listing
  • Inbox ID mismatch. Check the IDs from server instructions, not guessed values.
  • Search terms too narrow. Try broader terms or a longer time range.

Need more detail? Enable debug logging:

bash
LOG_LEVEL=debug npx [email protected]

Development

bash
git clone https://github.com/drewburchfield/help-scout-mcp-server.gitcd help-scout-mcp-servernpm install && npm run buildnpm start
bash
npm test           # Run testsnpm run type-check # TypeScript validationnpm run lint       # Lintingnpm run dev        # Development server with auto-reload

Contributions welcome. Please ensure tests, type checking, and linting pass before submitting a PR.

Support

License

MIT License - see LICENSE for details.

来源:README.md,提交 56c3434

工具

0
SourceWeft 尚未在沙箱中启动过该软件包,因此其工具未知。

版本历史

1
  1. v1.7.0最新Sep 16, 2026

评分

已满足适用检查项的 50%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通25 分

    请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。

  • 已声明认证方式10 分

    请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。

  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分