Tabelog Restaurant Search

Step 0 — Clarify Parameters

Determine the following from the user's request. Ask only for what's missing.

Choosing sort_by: Use ranking by default — it surfaces the most acclaimed restaurants based on Tabelog's composite score, which is a more reliable quality signal. Switch to review_count only when the user explicitly asks, e.g. "依評論數", "by most reviews", "口コミが多い順".

Location translation — convert the user's input to Japanese before searching, since Tabelog search is optimised for Japanese text:

Adding 「駅」 to station names gives geographically precise results. Prefer 目黒駅 over 目黒.

Step 1 — Open Tabelog

Navigate to https://tabelog.com.

⚠️ Language selector popup: On a fresh session, Tabelog shows a language selection popup at the bottom of the page. Click 「日本語」 to dismiss it before proceeding. If skipped, interactions may fail.

Confirm the Japanese homepage has loaded before moving on.

Step 2 — Fill the Area Field (UI interaction required)

Do NOT fill the area field by setting its value via JavaScript, and do NOT submit the form via JavaScript — both bypass Tabelog's area validation. The area will show 「全国」 (nationwide) instead of the target station, and JS form submission triggers a Yahoo CAPTCHA.

The correct sequence is:

1. Click the area input (name="sa", placeholder エリア・駅 [例:銀座、渋谷])
2. Type the location name (e.g. 上野駅) — this triggers an autocomplete dropdown
3. Wait ~1–2 seconds for the autocomplete suggestion to appear
4. Click the autocomplete suggestion matching the station name — this registers the area properly. The suggestion element type varies: it may be <button>, <li>, or <div>. Match by text content, not by element type.
5. Fill the keyword field (name="sk") with the cuisine type if provided (skip if none)
6. Click the 検索 button — use ID js-global-search-btn for reliable targeting

Without step 4, the search defaults to nationwide results even if the input shows the station name.

Step 3 — Verify Search Results Page

The page title after a successful search starts with the station name:

If the title shows 「全国のお店」 or does not start with the station name, the area filter didn't apply — repeat Step 2.

Step 4 — Switch Sort Tab

Recommended — navigate directly by URL (more reliable than clicking):

// Get the target tab URL from the page and navigate to it
Array.from(document.querySelectorAll('a'))
  .filter(function(a) { return a.textContent.trim() === 'ランキング'; })
  .map(function(a) { return a.href; })
  .join('|')

Take the returned URL and navigate to it directly with playwright-cli goto <url> or the equivalent navigate action.

Fallback — click the tab if JS returns empty: Find the tab link by text ランキング in the page state and click it. The four tabs in order: 標準 → ランキング → 口コミが多い順 → ニューオープン.

Confirm success: page title should include 「ランキング」.

Step 5 — Extract Results

Scroll down ~1000px to ensure all cards are rendered, then extract with JavaScript.

Confirmed CSS class names (as of 2025–2026)

Extraction script

Important for playwright-cli / agent-browser: console.log() output is not captured — only the final expression value is returned. Avoid complex multi-statement scripts; use simple single-expression evals or pipe-delimited string building to stay reliable.

// Run this as a single eval — returns pipe-delimited rows, one restaurant per line
(function() {
  var links = document.querySelectorAll('a[data-ranking]');
  var out = [];
  for (var i = 0; i < Math.min(10, links.length); i++) {
    var l = links[i];
    var card = l.closest('.list-rst--ranking');
    var score   = card.querySelector('.c-rating__val--strong')?.textContent?.trim() || '';
    var reviews = card.querySelector('.list-rst__rvw-count-num')?.textContent?.trim() || '';
    var area    = (card.querySelector('.list-rst__area-genre')?.textContent || '').replace(/\s+/g, ' ').trim().slice(0, 60);
    var badge   = card.querySelector('[class*="hyakumeiten"]') ? '百名店' : '';
    out.push(l.dataset.ranking + '|' + l.textContent.trim() + '|' + score + '|' + reviews + '|' + area + '|' + badge + '|' + l.href);
  }
  return out.join('\n');
})()

Parse each line by splitting on |: rank | name | score | reviewCount | areaGenre | badge | url

If the script returns empty, fall back to screenshots and read the cards visually.

Step 6 — Open Detail Pages (for top open_top_n results)

When open_top_n > 1 — use parallel subagents (recommended)

Spawn one subagent per restaurant in the same message so they run concurrently. Each subagent gets an isolated browser session via --session r1, --session r2, etc.

Subagent prompt template (repeat for each restaurant, changing session name and URL):

Use playwright-cli to extract restaurant details from Tabelog.
Run in sequence:
1. playwright-cli -s=r<N> open "<url>"
2. (wait 2 seconds)
3. playwright-cli -s=r<N> eval "document.querySelector('.p-rst-intro__txt, .pr-comment, .rstdtl-top__pr-comment-body')?.textContent?.trim()?.slice(0, 200)"
4. playwright-cli -s=r<N> eval "(function(){ var f={}; document.querySelectorAll('.rstinfo-table tr').forEach(function(r){ var l=r.querySelector('th')?.textContent?.trim(); var v=r.querySelector('td')?.textContent?.replace(/\s+/g,' ').trim().slice(0,100); if(l&&v) f[l]=v; }); return JSON.stringify(f); })()"
5. playwright-cli -s=r<N> close
Return all output from steps 3 and 4.

Note: The want array filter with Japanese strings can be unreliable in subagents. Extract the full info table (step 4 above) and let the subagent return everything — then filter to the fields you need when composing the final output.

When open_top_n = 1 — navigate directly in main session

// PR intro
document.querySelector('.p-rst-intro__txt, .pr-comment, .rstdtl-top__pr-comment-body')?.textContent?.trim()?.slice(0, 200)

// Info table — full extraction, filter afterwards
(function() {
  var f = {};
  document.querySelectorAll('.rstinfo-table tr').forEach(function(row) {
    var label = row.querySelector('th')?.textContent?.trim();
    var value = row.querySelector('td')?.textContent?.replace(/\s+/g, ' ').trim().slice(0, 100);
    if (label && value) f[label] = value;
  });
  return JSON.stringify(f);
})()

Some restaurants (especially small casual shops) have no PR text — skip gracefully and use the info table only.

Error Handling

Output Format

Present results in Traditional Chinese:

📍 目黒駅 附近拉麵排名（食べログ ランキング）

🥇 第1名：えーちゃん食堂
   評分：3.80 ｜ 評論數：1,052則
   距離：目黒駅 680m（不動前駅 659m）
   類型：ラーメン・つけ麺
   🏅 ラーメン TOKYO 百名店 2025

🥈 第2名：支那ソバ かづ屋
   評分：3.77 ｜ 評論數：1,763則
   ...

（最多列出前5~10名）

💡 資料來源：食べログ（tabelog.com）
⚠️  評分說明：3.5以上良好，3.8以上優秀，4.0以上為頂級名店

Highlight the 百名店 badge when present — it means the restaurant is on Tabelog's annual Top 100 list and is a strong quality signal worth calling out.

Scoring Reference