feat: 添加 Exa AI 搜索适配器

- 新增 ExaSearchAdapter，基于 MCP 协议调用 Exa 搜索 API
- WebSearchTool 支持 num_results、livecrawl、search_type、context_max_characters 等高级选项
- 非 Anthropic 官方 base URL 时默认使用 Exa 适配器

packages/builtin-tools/src/tools/WebSearchTool/adapters/exaAdapter.ts

@@ -0,0 +1,200 @@
+/**
+ * Exa AI-based search adapter — uses MCP protocol to call Exa's web search API.
+ *
+ * Ported from kilocode's production-validated implementation (mcp-exa.ts + websearch.ts).
+ * Key improvements over previous version:
+ *   - Passes through numResults/livecrawl/type/contextMaxCharacters from options
+ *   - Cleaner SSE parsing matching kilocode's approach
+ *   - Proper content snippet extraction from Exa responses
+ */
+
+import axios from 'axios'
+import { AbortError } from 'src/utils/errors.js'
+import type { SearchResult, SearchOptions, WebSearchAdapter } from './types.js'
+
+const EXA_MCP_URL = 'https://mcp.exa.ai/mcp'
+const FETCH_TIMEOUT_MS = 25_000
+
+export class ExaSearchAdapter implements WebSearchAdapter {
+  async search(
+    query: string,
+    options: SearchOptions,
+  ): Promise<SearchResult[]> {
+    const { signal, onProgress, allowedDomains, blockedDomains } = options
+
+    if (signal?.aborted) {
+      throw new AbortError()
+    }
+
+    onProgress?.({ type: 'query_update', query })
+
+    const abortController = new AbortController()
+    if (signal) {
+      signal.addEventListener('abort', () => abortController.abort(), { once: true })
+    }
+
+    // Use options to derive search params — matches kilocode websearch.ts defaults
+    const numResults = options.numResults ?? 8
+    const livecrawl = options.livecrawl ?? 'fallback'
+    const searchType = options.searchType ?? 'auto'
+    const contextMaxCharacters = options.contextMaxCharacters ?? 10000
+
+    let responseText: string
+    try {
+      const response = await axios.post(
+        EXA_MCP_URL,
+        {
+          jsonrpc: '2.0',
+          id: 1,
+          method: 'tools/call',
+          params: {
+            name: 'web_search_exa',
+            arguments: {
+              query,
+              type: searchType,
+              numResults,
+              livecrawl,
+              contextMaxCharacters,
+            },
+          },
+        },
+        {
+          signal: abortController.signal,
+          timeout: FETCH_TIMEOUT_MS,
+          headers: {
+            'Content-Type': 'application/json',
+            Accept: 'application/json, text/event-stream',
+          },
+          responseType: 'text',
+        },
+      )
+      responseText = response.data as string
+    } catch (e) {
+      if (axios.isCancel(e) || abortController.signal.aborted) {
+        throw new AbortError()
+      }
+      throw e
+    }
+
+    if (abortController.signal.aborted) {
+      throw new AbortError()
+    }
+
+    const searchText = this.parseSse(responseText)
+
+    if (abortController.signal.aborted) {
+      throw new AbortError()
+    }
+
+    // Parse the Exa results from the text response
+    const results = this.parseResults(searchText)
+
+    // Client-side domain filtering
+    const filteredResults = results.filter((r) => {
+      if (!r.url) return false
+      try {
+        const hostname = new URL(r.url).hostname
+        if (allowedDomains?.length && !allowedDomains.some(d => hostname === d || hostname.endsWith('.' + d))) {
+          return false
+        }
+        if (blockedDomains?.length && blockedDomains.some(d => hostname === d || hostname.endsWith('.' + d))) {
+          return false
+        }
+      } catch {
+        return false
+      }
+      return true
+    })
+
+    onProgress?.({
+      type: 'search_results_received',
+      resultCount: filteredResults.length,
+      query,
+    })
+
+    return filteredResults
+  }
+
+  private parseSse(body: string): string | undefined {
+    // SSE format: lines starting with "data: " containing JSON
+    // Matches kilocode mcp-exa.ts parseSse implementation
+    for (const line of body.split('\n')) {
+      if (!line.startsWith('data: ')) continue
+      const data = line.substring(6).trim()
+      if (!data || data === '[DONE]' || data === 'null') continue
+
+      try {
+        const parsed = JSON.parse(data)
+        const content = parsed?.result?.content
+        if (Array.isArray(content) && content[0]?.text) {
+          return content[0].text
+        }
+      } catch {
+        // Continue to next line
+      }
+    }
+
+    // Fallback: try parsing as direct JSON response (non-SSE)
+    try {
+      const parsed = JSON.parse(body)
+      const content = parsed?.result?.content
+      if (Array.isArray(content) && content[0]?.text) {
+        return content[0].text
+      }
+    } catch {
+      // Not JSON
+    }
+
+    return undefined
+  }
+
+  private parseResults(text: string | undefined): SearchResult[] {
+    if (!text) return []
+
+    const results: SearchResult[] = []
+
+    // Exa returns structured text with "Title:", "URL:", and "Content:" fields
+    // separated by "---" between entries
+    const blocks = text.split(/\n---\n/g)
+
+    for (const block of blocks) {
+      const titleMatch = block.match(/^Title:\s*(.+)$/m)
+      const urlMatch = block.match(/^URL:\s*(https?:\/\/[^\s]+)$/m)
+      const contentMatch = block.match(/^Content:\s*([\s\S]+?)(?=\n(?:Title:|URL:|---)|$)/m)
+
+      if (urlMatch) {
+        results.push({
+          title: titleMatch?.[1]?.trim() ?? urlMatch[1],
+          url: urlMatch[1].trim(),
+          snippet: contentMatch?.[1]?.trim().slice(0, 300),
+        })
+      }
+    }
+
+    // Fallback: markdown links
+    if (results.length === 0) {
+      const markdownLinkRegex = /\[([^\]]+)\]\((https?:\/\/[^\)]+)\)/g
+      let match: RegExpExecArray | null
+      while ((match = markdownLinkRegex.exec(text)) !== null) {
+        results.push({
+          title: match[1].trim(),
+          url: match[2].trim(),
+        })
+      }
+    }
+
+    // Fallback: plain URLs
+    if (results.length === 0) {
+      const urlRegex = /^https?:\/\/[^\s<>"\]]+/gm
+      let match: RegExpExecArray | null
+      while ((match = urlRegex.exec(text)) !== null) {
+        results.push({
+          title: match[0],
+          url: match[0],
+        })
+      }
+    }
+
+    return results
+  }
+}

packages/builtin-tools/src/tools/WebSearchTool/adapters/index.ts

@@ -7,6 +7,7 @@ import { isFirstPartyAnthropicBaseUrl } from 'src/utils/model/providers.js'
 import { ApiSearchAdapter } from './apiAdapter.js'
 import { BingSearchAdapter } from './bingAdapter.js'
 import { BraveSearchAdapter } from './braveAdapter.js'
+import { ExaSearchAdapter } from './exaAdapter.js'
 import type { WebSearchAdapter } from './types.js'
 
 export type {
@@ -17,16 +18,16 @@ export type {
 } from './types.js'
 
 let cachedAdapter: WebSearchAdapter | null = null
-let cachedAdapterKey: 'api' | 'bing' | 'brave' | null = null
+let cachedAdapterKey: 'api' | 'bing' | 'brave' | 'exa' | null = null
 
 export function createAdapter(): WebSearchAdapter {
   const envAdapter = process.env.WEB_SEARCH_ADAPTER
   const adapterKey =
-    envAdapter === 'api' || envAdapter === 'bing' || envAdapter === 'brave'
+    envAdapter === 'api' || envAdapter === 'bing' || envAdapter === 'brave' || envAdapter === 'exa'
       ? envAdapter
       : isFirstPartyAnthropicBaseUrl()
         ? 'api'
-        : 'bing'
+        : 'exa'
 
   if (cachedAdapter && cachedAdapterKey === adapterKey) return cachedAdapter
 
@@ -36,9 +37,14 @@ export function createAdapter(): WebSearchAdapter {
     return cachedAdapter
   }
   if (adapterKey === 'brave') {
-	  cachedAdapter = new BraveSearchAdapter()
-	  cachedAdapterKey = 'brave'
-	  return cachedAdapter
+    cachedAdapter = new BraveSearchAdapter()
+    cachedAdapterKey = 'brave'
+    return cachedAdapter
+  }
+  if (adapterKey === 'exa') {
+    cachedAdapter = new ExaSearchAdapter()
+    cachedAdapterKey = 'exa'
+    return cachedAdapter
   }
 
   cachedAdapter = new BingSearchAdapter()

packages/builtin-tools/src/tools/WebSearchTool/adapters/types.ts

@@ -9,6 +9,14 @@ export interface SearchOptions {
   blockedDomains?: string[]
   signal?: AbortSignal
   onProgress?: (progress: SearchProgress) => void
+  /** Number of search results to return (default: 8) */
+  numResults?: number
+  /** Live crawl mode (default: 'fallback') */
+  livecrawl?: 'fallback' | 'preferred'
+  /** Search type (default: 'auto') */
+  searchType?: 'auto' | 'fast' | 'deep'
+  /** Maximum characters for context string (default: 10000) */
+  contextMaxCharacters?: number
 }
 
 export interface SearchProgress {