Export Public Comments on Chosen TikTok Videos to CSV

TikTok Comments Export turns a few video ids or links into one CSV of the public comments Desearch returned, plus a short summary: comments per page, a simple question / complaint / praise tag, and the top 10 by likes.

How this differs

Voice of customer analysis from Instagram and TikTok comments starts from a brand's public usernames, reads the latest posts on both apps, and sorts comments into groups. This script does not read Instagram and does not look up a profile. You pass TikTok video links you already have, and it writes a CSV of those comment pages.

The problem

The comments you care about are often on a few specific videos: your own, a partner's, or one post in a topic you track. Opening each video and copying comments loses likes and time. A sheet you can sort is the useful file.

What the app does

• Takes TikTok video ids or links (the long number in the video URL).
• Calls GET /desearch/tiktok/comments/{video_id} once per page.
• When the list has no cursor field, sends the running offset as cursor and dedupes anyway.
• Builds a comment_key from a short hash of commenter id, time and text when the response has no comment id.
• Shows commenters as Viewer N. Mentions in the text become [viewer] and links become [link].
• Tags each comment with a keyword rule: question, complaint, praise, or blank.
• Writes the CSV and a summary. --from-raw rebuilds the file from saved JSON.

How it works

1. Comments. GET /desearch/tiktok/comments/{video_id}?count=... for each video and page.
2. Dedupe. Same comment_key on a later page is dropped.
3. Tag. A question mark or a question word, a complaint word, or a praise word. The tag is a starting point for reading, not a classifier.
4. Sort in the sheet. The order inside a page is not by likes or by time.

The code

The comment key and the tag:

def comment_key(c):
    """Comment id if returned; the API returned none in testing, so fall back to a hash of author id + time + text."""
    if c.get("id"): return str(c["id"])
    a = (c.get("author") or {}).get("id") or ""
    return "h" + hashlib.sha256(f"{a}|{c.get('createTime')}|{c.get('text')}".encode()).hexdigest()[:12]


def tag(text):
    low = text.lower()
    return next((n for n, rx in TAGS if re.search(rx, low)), "")

The live calls:

def run_live(a):
    import requests
    key = os.environ.get("DESEARCH_API_KEY") or open(os.path.expanduser(os.environ["DESEARCH_KEY_FILE"])).read().strip()
    s = requests.Session(); s.headers["Authorization"] = key
    raw = {"count": a.count, "pages": a.pages, "started": now(), "calls": []}
    for v in a.videos:
        vid, cursor = video_id(v), None
        for i in range(a.pages):
            params = {"count": a.count}
            if cursor: params["cursor"] = cursor
            rec = call(s, vid, params, f"{vid[-6:]} p{i+1}"); raw["calls"].append(rec)
            if not comments(rec): break
            cursor, rec["cursor_kind"] = next_cursor(rec, a.count * (i + 1))
    if a.empty_check:
        rec = call(s, a.empty_check, {"count": 5}, "empty-result check"); rec["probe"] = True; raw["calls"].append(rec)
    os.makedirs("evidence", exist_ok=True)
    json.dump(raw, open(f"evidence/raw-{a.label}.json", "w"), indent=1, default=str)
    return raw

The key comes from the environment and is sent as the Authorization header. Video links in the CSV are rebuilt from the id, with no account name in them.

Endpoints

• GET /desearch/tiktok/comments/{video_id}

Example output

A test run on 2026-10-16 used 2 videos, 10 comments per page, and 3 pages each. It wrote a CSV of unique comments in under 10 seconds. Commenters are Viewer N. reply_count is blank when the response has no reply data. You get the public comments returned for those pages, not every comment on a video.

Run it yourself

The public examples folder linked above includes tiktok_comments.py and a README. Install its requirements, set DESEARCH_API_KEY from console.desearch.ai, and use the run command in that README.

What it costs

Each response records X-Desearch-Cost-Usd. The sample write-up does not publish a dollar total. Keep --count and --pages small on the first run and read the header before you scale. See pricing.

Ideas to extend it

• Sort the summary by tag before you read the top likes.
• Add the video caption as a column from a saved post export.
• Keep a file of comment keys you have already read.

{
  "mcpServers": {
    "desearch": {
      "url": "https://mcp.desearch.ai/mcp",
      "headers": {
        "Authorization": "Bearer <DESEARCH_API_KEY>"
      }
    }
  }
}

Use Desearch to read public comments on the TikTok video links I give you. Page them, drop duplicate comments, and write a CSV with likes, time and a simple tag. Show commenters as Viewer 1, Viewer 2 and so on. Do not invent comments.