feat: KOC 资源导入后自动爬取主页资料并调用 LLM 打标签

- 新增 structured_text_process 打标客户端（文本构造、标签解析清洗、MCP 调用）
- 写库后后台打标（tagUntaggedAccounts），单条新增 / 批量导入 / 每日补全三个入口统一
- 标签只补空，UPDATE 带 tags 为空守卫，不覆盖已有标签
- 新增 AI_TAGGING_TOOL 配置项（留空关闭），更新配置示例与文档
- 新增 16 个打标测试用例，全量 131 个测试通过

lib/mcp-collection-client.ts

@@ -1484,3 +1484,154 @@ export async function resolveXhsProfileDetailsFromMcp(
     ? lastError
     : new Error("账号主页数据采集服务暂时不可用");
 }
+
+export type AccountTaggingInput = {
+  platform: string;
+  nickname: string;
+  bio: string;
+  gender: string;
+  ipLocation: string;
+  followers: number | null;
+  providerTags: string[];
+  recentNoteTitles: string[];
+};
+
+export type AccountTaggingEnv = {
+  AI_TAGGING_TOOL?: string;
+};
+
+export function buildAccountTagText(input: AccountTaggingInput): string {
+  const facts: string[] = [`平台：${input.platform.trim() || "小红书"}`];
+  if (input.nickname.trim()) facts.push(`昵称：${input.nickname.trim()}`);
+  if (input.bio.trim()) facts.push(`简介：${input.bio.trim()}`);
+  if (input.gender.trim()) facts.push(`性别：${input.gender.trim()}`);
+  if (input.ipLocation.trim()) {
+    facts.push(`IP属地：${input.ipLocation.trim()}`);
+  }
+  if (input.followers !== null && input.followers > 0) {
+    facts.push(`粉丝数：${input.followers}`);
+  }
+  if (input.providerTags.length > 0) {
+    facts.push(`平台标签：${input.providerTags.slice(0, 5).join("、")}`);
+  }
+  if (input.recentNoteTitles.length > 0) {
+    const titles = input.recentNoteTitles
+      .slice(0, 20)
+      .map((title) => `《${title}》`)
+      .join("");
+    facts.push(`最近笔记标题：${titles}`);
+  }
+  return facts.join("\n");
+}
+
+function sanitizeTagList(items: unknown[]): string[] {
+  const tags: string[] = [];
+  for (const item of items) {
+    if (typeof item !== "string") continue;
+    const tag = item.trim().replace(/^#+|#+$/g, "").trim();
+    if (!tag || tag.length > 12) continue;
+    if (!tags.includes(tag)) tags.push(tag);
+  }
+  return tags.slice(0, 5);
+}
+
+export function parseAccountTags(text: string): string[] {
+  const stripped = text
+    .trim()
+    .replace(/^```(?:json)?\s*/i, "")
+    .replace(/\s*```\s*$/, "");
+  const candidates = [stripped];
+  const arrayMatch = stripped.match(/\[[\s\S]*\]/);
+  if (arrayMatch) candidates.push(arrayMatch[0]);
+  for (const candidate of candidates) {
+    let parsed: unknown;
+    try {
+      parsed = JSON.parse(candidate);
+    } catch {
+      continue;
+    }
+    if (!Array.isArray(parsed)) continue;
+    const tags = sanitizeTagList(parsed);
+    if (tags.length > 0) return tags;
+  }
+  return [];
+}
+
+function parseTagsFromToolPayload(payload: unknown, depth = 0): string[] {
+  if (typeof payload === "string") return parseAccountTags(payload);
+  if (Array.isArray(payload)) {
+    const direct = sanitizeTagList(payload);
+    if (direct.length > 0) return direct;
+    if (depth < 2) {
+      for (const item of payload) {
+        const tags = parseTagsFromToolPayload(item, depth + 1);
+        if (tags.length > 0) return tags;
+      }
+    }
+    return [];
+  }
+  const record = asRecord(payload);
+  if (record && depth < 2) {
+    for (const key of [
+      "tags",
+      "labels",
+      "result",
+      "data",
+      "response",
+      "output",
+      "text",
+      "content",
+      "message",
+    ]) {
+      if (key in record) {
+        const tags = parseTagsFromToolPayload(record[key], depth + 1);
+        if (tags.length > 0) return tags;
+      }
+    }
+  }
+  return [];
+}
+
+export async function generateAccountTagsFromMcp(
+  input: AccountTaggingInput,
+  config: CollectionMcpConfig,
+  taggingEnv: AccountTaggingEnv,
+  fetchImpl: typeof fetch = fetch,
+): Promise<string[]> {
+  const toolName = String(taggingEnv.AI_TAGGING_TOOL ?? "").trim();
+  if (!toolName || !config.endpoint.trim()) return [];
+  try {
+    const endpoint = buildMcpUrl(config);
+    const timeoutMs = Math.max(5_000, config.timeoutMs ?? 60_000);
+    const sessionId = await createMcpSession(fetchImpl, endpoint, timeoutMs);
+    const result = await invokeMcpTool(
+      fetchImpl,
+      endpoint,
+      sessionId,
+      timeoutMs,
+      toolName,
+      {
+        objective: "为KOC账号生成运营标签",
+        text: buildAccountTagText(input),
+        rules: [
+          "最多生成 5 个标签",
+          "每个标签不超过 8 个字",
+          "从内容领域、人设风格、受众画像等维度概括",
+          "只依据文本中给到的资料，不要编造",
+        ],
+        output_schema: {
+          type: "object",
+          properties: {
+            tags: { type: "array", items: { type: "string" } },
+          },
+          required: ["tags"],
+          additionalProperties: false,
+        },
+      },
+    );
+    if (result.isError) return [];
+    return parseTagsFromToolPayload(result.payload);
+  } catch {
+    return [];
+  }
+}