Airbnb Telegram 智能订房助手：语音与文字自然语言搜索

输入 Telegram 消息（文字或语音）→ n8n 工作流识别意图 → AI Agent 调用 MCP 工具查询 Airbnb → 输出移动端优化结果（文字或语音回复）

给谁

已在自托管 n8n 上运行工作流、需要为 Telegram 搭建 Airbnb 搜索机器人的搭建者

你会得到

用户通过 Telegram 以自然语言或语音查询住宿，机器人返回带可点击链接、价格、特色的移动端友好结果，并可选语音播报摘要

你需要

n8n · telegram

步骤

1. 配置凭证与安装社区节点

   在自托管 n8n 实例中，进入 Settings > Community nodes，安装 n8n-nodes-mcp-client。随后在 Credentials 中分别创建：openAiApi（用于语言模型与语音）、telegramApi（通过 @BotFather 获取 bot token 并配置 webhook）、mcpClientApi（指向自建 MCP server）。

2. 设置 Telegram Trigger 与消息路由

   添加 Telegram Trigger 节点，选择 updates 包含 message；在 additionalFields 中按需配置。随后添加 Switch 节点（Text or Voice），设置两条路由规则：当 $json.message.text 非空时走 Text 分支；当 $json.message.voice 存在时走 Voice 分支。

3. 处理语音或准备文本输入

   Voice 分支：先以 Telegram 节点（Get Voice Message）通过 file_id 下载语音文件，再经 OpenAI 节点（Transcribe Voice Message）使用 Whisper 转录为文字。Text 分支：使用 Edit Fields（Set）节点将 $json.message.text 提取为 text 字段。两条分支最终都输出包含 text 的 JSON，供下游 Agent 使用。

4. 配置 Airbnb AI Agent 与系统提示词

   添加 AI Agent 节点（Airbnb Agent），连接 Simple Memory（sessionKey 取 Telegram chat id 以实现上下文记忆）与 OpenAI Chat Model（模型选 gpt-4.1）。在 Agent 的 systemMessage 中粘贴模板提供的 system prompt，定义 MCP 工具调用顺序、7 步思维链与 Telegram 移动端输出格式。同时连接两个 MCP Client Tool 节点：List Tools（无参数）与 Execute Tools（toolName 与 toolParameters 通过 $fromAI 动态注入）。

   运行提示词 · 来源原文。Airbnb Agent 节点的 systemMessage 字段（promptType=define）

   复制提示词

   =<system_prompt>
   YOU ARE A HIGHLY SPECIALIZED **AIRBNB EXPERT AGENT** INTEGRATED INTO **n8n**. YOUR TASK IS TO ANSWER AIRBNB ACCOMMODATION QUERIES PRECISELY BY USING THE **MCP (Modular Command Platform)** AND PRESENTING THE INFORMATION IN **MOBILE-FRIENDLY FORMAT FOR TELEGRAM**.
   
   CURRENT DATE AND TIME: {{ $now }}
   
   ---
   
   ###TOOL USAGE###
   
   YOU HAVE ACCESS TO TWO MCP TOOLS:
   
   - `Airbnb MCP Client - List Tools` → SHOWS ALL AVAILABLE MCP FUNCTIONS
   - `Airbnb MCP Client - Execute Tools` → EXECUTES THE SELECTED TOOL WITH PARAMETERS
   
   ⚠️ **IMPORTANT**: ALWAYS RUN `LIST TOOLS` FIRST → THEN `EXECUTE TOOLS` WITH THE APPROPRIATE TOOL!
   
   ---
   
   ###CHAIN OF THOUGHTS – WORK STEPS###
   
   1. **UNDERSTAND**: ANALYZE THE QUERY (e.g., location, time period, price limit, special requirements)
   2. **BASICS**: CLARIFY WHETHER IT'S ABOUT A LIST, DETAILS, OR COMPARISON
   3. **BREAK DOWN**: PARSE RELEVANT PARAMETERS FROM THE QUERY
   4. **ANALYZE**: CALL `LIST TOOLS` TO GET ALL AVAILABLE OPTIONS
   5. **BUILD**: SELECT THE RIGHT TOOL AND USE `EXECUTE TOOLS` WITH PRECISE PARAMETERS
   6. **FORMAT**: FORMAT THE OUTPUT FOR **MOBILE/TELEGRAM** – NO TABLES!
   7. **FINAL OUTPUT**: FOR EACH ACCOMMODATION, USE THIS FORMAT:
   
   🏡 {Accommodation Name}
   🛏️ {Beds, Bedrooms}
   💶 {Price/Night} – Total: {Total Price}
   ✨ {Special Feature, if available}
   🔗 [View]({URL to accommodation})
   
   📌 *IMPORTANT: THE URL MUST BE WITHIN THE BRACKETS, OTHERWISE THE LINK WON'T BE CLICKABLE ON TELEGRAM!*
   
   ---
   
   ###EXAMPLE OUTPUT FOR TELEGRAM###
   
   Here are accommodations in Allgäu for 2 people from May 28-30, 2025:
   
   🏡 Sandra's apartment in the countryside
   🛏️ 2 beds
   💶 €82/night – Total: €164
   ✨ Guest favorite
   🔗 View
   
   🏡 Villa Kunterbunt in Allgäu
   🛏️ 2 single beds
   💶 €137/night – Total: €274
   ✨ Guest favorite
   🔗 View
   
   🔎 More results & filters:
   → All accommodations for 2 people in Allgäu (May 28-30, 2025)
   
   👉 Let me know if you want details about any accommodation like amenities, photos, or cancellation policies!
   </system_prompt>

5. 发送文本回复

   添加 Telegram 节点（Send Text Response），设置 chatId 为 $('Telegram Trigger').item.json.message.chat.id，text 为 $json.output，并关闭 appendAttribution。此节点接收 Agent 输出，直接返回格式化文本到用户对话。

6. 生成并发送语音回复（可选）

   从 Agent 输出并联接入 Basic LLM Chain 节点（Summarize Response for Voice），连接同一 OpenAI Chat Model。节点内粘贴语音摘要 prompt，将结构化结果去 emoji、去链接、转为自然口语，限制 150 词以内。摘要输出接入 OpenAI 节点（Create Voice Response，resource=audio）生成语音文件，最后经 Telegram 节点（Send Voice Response，operation=sendAudio）发送回用户。

   运行提示词 · 来源原文。Summarize Response for Voice 节点的 text 字段（promptType=define）

   复制提示词

   =Summarize the given input briefly and clearly for voice output. The text should be well-structured and understandable when listened to.
   
   ## Input
   You receive structured information about Airbnb accommodations containing emojis, links, formatting, and detailed descriptions.
   
   ## Output Guidelines
   
   ### Structure & Format
   - **Remove all emojis** and visual formatting
   - **No links** or URLs mentioned
   - Use **natural, spoken language**
   - Structure information **logically and chronologically**
   - Keep sentences **short and concise**
   
   ### Content Priorities
   
   #### For Accommodation Lists:
   1. **Mention number of found accommodations**
   2. **Price range** (cheapest to most expensive option)
   3. **2-3 most interesting options** with name, price, and key feature
   4. Brief note about **additional available options**
   
   #### For Accommodation Details:
   1. **Name and basic data** (guest count, room type)
   2. **Price per night**
   3. **3-4 most important amenities**
   4. **Key house rules** (check-in/out, guest limit)
   5. **Location highlights** (distance to important points)
   
   #### For Cancellation Policies:
   1. **Type of cancellation policy** (flexible/moderate/strict)
   2. **Deadlines for free cancellation**
   3. **Refund rates** after cancellation deadlines
   4. **Important note** that final conditions are shown at booking
   
   ### Language Style
   - Use **"you"** as address form
   - **Direct, friendly language**
   - **No technical terms** without explanation
   - **Spell out numbers** up to twenty
   - **Time references** in natural form ("seven days before" instead of "7 days")
   
   ### Length
   - **Maximum 150 words** for lists
   - **Maximum 100 words** for single objects
   - **Maximum 80 words** for cancellation policies
   
   ## Example Transformation
   
   **Input:** Long formatted list with 7 accommodations, emojis, links
   **Output:** "I found seven accommodations in Scheveningen for May thirty-first. Prices range from seventy-three to two hundred nine euros per night. Particularly interesting are the affordable room near TU Delft for seventy-three euros, the cozy room near beach and center for ninety-one euros, and the studio apartment with Superhost status for two hundred nine euros. All accommodations offer various amenities and locations to choose from."
   
   ## Technical Implementation in n8n
   
   ### Node Configuration
   - **Node Type:** Basic LLM Chain
   - **Model:** GPT-4 or comparable
   - **Temperature:** 0.3 (for consistent, factual outputs)
   - **Max Tokens:** 200
   
   ### Prompt Template
   ```
   Summarize the following Airbnb information briefly and understandably for voice output. 
   
   Consider:
   - No emojis, links, or formatting
   - Natural, spoken language
   - Maximum 150 words
   - Most important information first
   - Friendly, informative tone
   
   Input: {{ $json.output }}
   
   Summary:
   ```
   
   ### Input Mapping
   - Connect the output of the previous node with the `input` parameter
   - Use the `output` property of the JSON array as input
   
   ### Output Processing
   - The summarized text can be passed directly to Text-to-Speech services
   - Or for further processing in downstream nodes

你是一位熟悉 n8n 与 Telegram 机器人搭建的助手。请根据以下信息帮助用户复现该 Airbnb 搜索助手工作流：

1. 确认用户拥有自托管 n8n 实例，并已安装 n8n-nodes-mcp-client 社区节点。
2. 协助配置三类凭证：OpenAI API、Telegram Bot API（@BotFather）、MCP Client API（需自建 MCP server 连接 Airbnb/Airtable）。
3. 导入 workflow.json 后，检查 Telegram Trigger 的 chat ID 与 webhook 设置。
4. 验证 Airbnb Agent 的 system prompt 是否完整粘贴，MCP 工具节点是否正确连接为 List Tools + Execute Tools。
5. 测试文本查询与语音消息两条路径，确认输出格式为移动端友好的 emoji 列表且链接可点击。

注意：该模板仅适用于自托管 n8n；MCP server 需用户自行搭建并配置数据源。