# Call Tree - Core Methods
## Overview
This document describes the call tree for the two core methods in the Bing CN MCP server.
---
## 1. searchBing() - 必应搜索方法
### Call Flow
```
MCP Client Request
↓
server.tool("bing_search", ...) [line 584]
↓
async ({ query, num_results }) => { ... } [line 591]
↓
await searchBing(query, num_results) [line 594]
↓
┌─────────────────────────────────────────────────────────┐
│ searchBing(query: string, numResults: number) [line 33] │
└─────────────────────────────────────────────────────────┘
│
├─→ Build search URL [line 36]
│ └─→ encodeURIComponent(query)
│
├─→ Set request headers [line 40-52]
│ └─→ USER_AGENT, Accept, Accept-Language, etc.
│
├─→ axios.get(searchUrl, { headers, timeout }) [line 55]
│ └─→ HTTP GET request to Bing CN
│
├─→ cheerio.load(response.data) [line 67]
│ └─→ Parse HTML response
│
├─→ Extract results using multiple selectors [line 83-303]
│ ├─→ Try '#b_results > li.b_algo'
│ ├─→ Try '#b_results > li.b_ans'
│ ├─→ Try '#b_results > li:not(.b_ad)...'
│ ├─→ Try '#b_topw > li.b_algo'
│ └─→ Try '#b_topw > li.b_ans'
│
├─→ For each result element [line 93]
│ ├─→ Extract title [line 108-178]
│ │ ├─→ Method 1: h2 a
│ │ ├─→ Method 2: .b_tpcn .tptt
│ │ ├─→ Method 3: .b_title a, a.tilk
│ │ └─→ Method 4: h2 text
│ │
│ ├─→ Extract link [line 109-178]
│ │ ├─→ From h2 a href
│ │ ├─→ From .tilk href/redirecturl/data-h
│ │ └─→ Fix relative URLs [line 195-210]
│ │
│ ├─→ Extract snippet [line 180-200]
│ │ ├─→ From .b_caption p
│ │ ├─→ From .b_snippet, .b_lineclamp2
│ │ └─→ From element text
│ │
│ └─→ Save to searchResults Map [line 290]
│ └─→ searchResults.set(id, result)
│
├─→ Fallback: Extract from links [line 305-371]
│ └─→ If no results found with selectors
│
└─→ Return results array [line 375]
└─→ Return SearchResult[]
│
↓
Return JSON stringified results [line 600]
↓
MCP Response to Client
```
### Key Dependencies
- **axios**: HTTP client for fetching Bing search page
- **cheerio**: HTML parsing and DOM manipulation
- **searchResults Map**: Global storage for results (used by fetchWebpageContent)
---
## 2. fetchWebpageContent() - 网页内容获取方法
### Call Flow
```
MCP Client Request
↓
server.tool("fetch_webpage", ...) [line 619]
↓
async ({ result_id }) => { ... } [line 625]
↓
await fetchWebpageContent(result_id) [line 628]
↓
┌──────────────────────────────────────────────────────────┐
│ fetchWebpageContent(resultId: string) [line 402] │
└──────────────────────────────────────────────────────────┘
│
├─→ Get result from searchResults Map [line 405]
│ └─→ searchResults.get(resultId)
│ └─→ Extract URL: result.link
│
├─→ Set request headers [line 414-428]
│ ├─→ User-Agent, Accept, Accept-Language
│ ├─→ Accept-Encoding: gzip, deflate, br
│ ├─→ Sec-Fetch-* headers
│ └─→ Dynamic Referer based on target URL [line 431-436]
│
├─→ Random delay (200-700ms) [line 440-441]
│ └─→ Avoid bot detection
│
├─→ axios.get(url, { headers, timeout, ... }) [line 443]
│ ├─→ responseType: 'arraybuffer'
│ ├─→ maxRedirects: 5
│ └─→ validateStatus: (status) => status < 500
│
├─→ Handle HTTP errors [line 453-460]
│ ├─→ 403 Forbidden → throw error
│ └─→ Other 4xx → throw error
│
├─→ Detect encoding [line 462-472]
│ ├─→ From Content-Type header
│ └─→ Fallback to UTF-8
│
├─→ Decode HTML content [line 474-481]
│ ├─→ TextDecoder(encoding)
│ └─→ Fallback to UTF-8 if fails
│
├─→ cheerio.load(html) [line 484]
│ └─→ Parse HTML
│
├─→ Remove unwanted elements [line 487]
│ └─→ script, style, iframe, nav, header, footer, ads, etc.
│
├─→ Extract main content [line 489-523]
│ ├─→ Try main selectors [line 492-500]
│ │ ├─→ main, article, .content, #content, etc.
│ │ └─→ Extract text from matched elements
│ │
│ ├─→ Try paragraph extraction [line 502-508]
│ │ └─→ $('p').each() → collect paragraphs
│ │
│ └─→ Fallback to body text [line 512-515]
│ └─→ $('body').text()
│
├─→ Clean content [line 518-521]
│ ├─→ Replace multiple spaces
│ └─→ Replace multiple newlines
│
├─→ Add page title [line 524-527]
│ └─→ $('title').text()
│
└─→ Truncate if too long [line 530-533]
└─→ Max 8000 characters
│
↓
Return content string [line 534]
↓
MCP Response to Client
```
### Key Dependencies
- **searchResults Map**: Get URL from previous search result
- **axios**: HTTP client for fetching webpage
- **cheerio**: HTML parsing and content extraction
- **TextDecoder**: Handle various character encodings
---
## Data Flow
### searchBing → fetchWebpageContent
```
1. searchBing() executes
└─→ Stores results in searchResults Map
└─→ Each result has: { id, title, link, snippet }
2. Client receives search results with IDs
└─→ Example: { id: "result_123", title: "...", link: "https://..." }
3. Client calls fetch_webpage with result_id
└─→ fetchWebpageContent("result_123")
4. fetchWebpageContent() retrieves URL from Map
└─→ const result = searchResults.get("result_123")
└─→ const url = result.link
5. Fetches and returns webpage content
```
---
## Error Handling
### searchBing()
- **Network errors**: Caught at line 376, returns error result
- **Parsing errors**: Logged, returns partial results or error result
- **No results**: Returns fallback result with search URL
### fetchWebpageContent()
- **Invalid resultId**: Throws error at line 407
- **403 Forbidden**: Thrown at line 455 with specific message
- **Other HTTP errors**: Thrown at line 459
- **Decoding errors**: Fallback to UTF-8 at line 479
- **All errors**: Re-thrown at line 558 with formatted message
---
## Global State
### searchResults Map (line 25)
- **Type**: `Map<string, SearchResult>`
- **Purpose**: Store search results for later retrieval by ID
- **Lifecycle**:
- Populated by `searchBing()`
- Read by `fetchWebpageContent()`
- Persists for the lifetime of the server process
---
## Summary
1. **searchBing()**:
- Fetches Bing search page
- Parses HTML with Cheerio
- Extracts results using multiple selector strategies
- Stores results in global Map
- Returns result array
2. **fetchWebpageContent()**:
- Retrieves URL from global Map using result ID
- Fetches webpage with browser-like headers
- Parses and extracts main content
- Cleans and truncates content
- Returns content string
Both methods are exposed as MCP tools and can be called independently by MCP clients.
