Check New X Posts for Keywords Since Your Last Look

X Keyword Monitor is a daily check for a few phrases on X. You keep a last-check time. Each run searches public posts, keeps only posts newer than that time, drops link-only posts, and prints the time to save for the next run. Authors are shown as a user. The rows are what Desearch returned, not a full count of the conversation.

How this differs

The Twitter complaints monitor pairs a brand with complaint words and sorts posts into themes. Export Twitter search results to CSV writes one date window to a file and does not remember the last run. This script stores a last-check time and, on the next run, keeps only posts created after it.

The problem

Typing the same X searches every morning means scrolling past posts you already saw. A keyword list without a cutoff fills up with yesterday. A community manager wants the new posts, a short summary, and a timestamp to paste in next time.

What the app does

• Reads keywords, a likes floor, a count and last_check (UTC) from config.json.
• Calls GET /twitter with the phrases OR'd inside parentheses, from the date of the last check through today.
• Keeps posts created after last_check.
• Drops posts whose text has none of the keywords, and posts that are only a link.
• Sorts what remains newest first, writes three summary lines, and prints the next last-check time.
• Shows each author as "a user".

How it works

1. Search. GET /twitter with sort=Latest, min_likes, count, start_date set to the date of last_check, and end_date set to today (YYYY-MM-DD).
2. Filter. The date on the request is a whole day, so the script drops posts that are not newer than the exact last-check time.
3. Write output/monitor.md and output/posts.csv. Copy Next LAST CHECK back into last_check.

The code

The query and the request:

def keyword_query(keywords):
    parts = ['"%s"' % term if " " in term else term for term in keywords]
    return "(" + " OR ".join(parts) + ")"


def x_params(cfg, now):
    last = parse_time(cfg["last_check"])
    return {
        "query": keyword_query(cfg["keywords"]),
        "sort": "Latest",
        "start_date": last.date().isoformat(),
        "end_date": now.date().isoformat(),
        "min_likes": int(cfg.get("min_likes", 5)),
        "count": int(cfg.get("count", 20)),
    }

Link-only posts are dropped after the response:

def link_only(text):
    stripped = re.sub(r"https?://\S+", " ", text or "")
    stripped = re.sub(r"www\.\S+", " ", stripped)
    return re.sub(r"[^\w]+", "", stripped) == ""

The live call:

def run_live(cfg, session=None, now=None):
    key = api_key()
    if session is None:
        import requests
        session = requests.Session()
    session.headers["Authorization"] = key
    if now is None:
        now = parse_time(cfg["now"]) if cfg.get("now") else dt.datetime.now(dt.timezone.utc)
    params = x_params(cfg, now)
    res = session.get(BASE + "/twitter", params=params, timeout=90)
    try:
        body = res.json()
    except ValueError:
        body = {"_text": getattr(res, "text", "")[:500]}
    raw = {
        "run_at": now.isoformat(timespec="seconds"),
        "now": now.isoformat(timespec="seconds"),
        "last_check": cfg["last_check"],
        "cfg": cfg,
        "x": {"status": res.status_code, "body": body, "params": params},
    }
    return raw, key

The key is sent only in the Authorization header and is not written to the output.

Endpoints

• GET /twitter

Example output

The 2026-10-15 sample used three home-coffee phrases, at least 5 likes, from 2026-10-14 11:35 UTC to 2026-10-15 11:35 UTC. 20 posts came back, 12 were newer than the last check, and 11 were kept. 1 link-only post was dropped. 8 posts were older than the last check. Authors in the public sample are shown as a user.

Run it yourself

The public examples folder linked above includes the script, config.json, a saved prompt and a README. Install its requirements, set DESEARCH_API_KEY from console.desearch.ai, and use the run command in that README. You can also paste the saved prompt into an MCP client pointed at https://mcp.desearch.ai/mcp.

What it costs

Each run is one GET /twitter call. The sample write-up does not publish a dollar total from X-Desearch-Cost-Usd. See pricing.

Ideas to extend it

• Run it on a schedule and append each CSV.
• Add a language note when you only want one language.
• Keep the dropped link-only rows in a second file.

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

Use Desearch to search public X posts for the phrases I give you, starting from the last-check time I set. Keep only posts newer than that time, drop link-only posts, and tell me the time to use next. Show authors as a user. Do not invent posts.