X/Twitter News Search

Search Twitter/X for recent tweets matching keywords and engagement thresholds. Returns raw tweet data as JSON — the calling agent handles scoring, formatting, and delivery.

Quick reference

curl -s -G "https://api.x.com/2/tweets/search/recent" \
  --data-urlencode 'query=("AI agent" OR "agentic AI") (lang:en) -is:reply -is:retweet' \
  --data-urlencode "max_results=30" \
  --data-urlencode "start_time=$(date -u -v-24H +%Y-%m-%dT%H:%M:%SZ)" \
  --data-urlencode "tweet.fields=created_at,public_metrics,author_id" \
  --data-urlencode "user.fields=username,name" \
  --data-urlencode "expansions=author_id" \
  --data-urlencode "sort_order=relevancy" \
  -H "Authorization: Bearer $TWITTER_BEARER_TOKEN" | jq .

Parameters

The caller provides these. Use defaults when not specified.

Procedure

1. Check API key

Verify $TWITTER_BEARER_TOKEN is set:

test -n "$TWITTER_BEARER_TOKEN" && echo "OK" || echo "MISSING"

If missing, tell the user:

1. Go to https://developer.x.com
2. Create a Project and attach an App
3. Copy the Bearer Token
4. Save it: hermes config set TWITTER_BEARER_TOKEN <token>

2. Build the query

Combine keywords with OR. Wrap multi-word phrases in quotes:

("keyword one" OR "keyword two" OR single) -is:reply -is:retweet

Add language filter: (lang:en OR lang:fr)

Available on pay-per-use: lang:, -is:reply, -is:retweet, from:, has:links, has:media

NOT available on pay-per-use: min_faves, min_retweets, since, until, -filter:replies — use API params and post-fetch filtering instead.

3. Call the API

Use start_time as an API parameter for the time window:

curl -s -G "https://api.x.com/2/tweets/search/recent" \
  --data-urlencode "query=QUERY_HERE" \
  --data-urlencode "max_results=50" \
  --data-urlencode "start_time=ISO_TIMESTAMP" \
  --data-urlencode "tweet.fields=created_at,public_metrics,author_id" \
  --data-urlencode "user.fields=username,name" \
  --data-urlencode "expansions=author_id" \
  --data-urlencode "sort_order=relevancy" \
  -H "Authorization: Bearer $TWITTER_BEARER_TOKEN"

Generate start_time:

• Linux: date -u -d "$N hours ago" +%Y-%m-%dT%H:%M:%SZ
• macOS: date -u -v-${N}H +%Y-%m-%dT%H:%M:%SZ

Fetch 3x max_results to have enough after engagement filtering.

4. Extract and filter

Parse with jq — see references/twitter-api.md for full response schema.

... | jq '[
  .data[] as $t |
  (.includes.users[] | select(.id == $t.author_id)) as $u |
  {
    id: $t.id,
    text: $t.text,
    author: $u.username,
    author_name: $u.name,
    likes: $t.public_metrics.like_count,
    retweets: $t.public_metrics.retweet_count,
    replies: $t.public_metrics.reply_count,
    views: $t.public_metrics.impression_count,
    created_at: $t.created_at,
    url: ("https://x.com/" + $u.username + "/status/" + $t.id)
  }
] | sort_by(-.likes)'

Filter by engagement:

... | jq --argjson ml 50 --argjson mr 5 '[.[] | select(.likes >= $ml and .retweets >= $mr)]'

Trim to max_results: | .[:30]

5. Return the data

Return the filtered array as JSON. Do not score, summarize, or format — just the raw tweet objects.

Pitfalls

• Empty results: Return empty array [], not an error. Some queries have no matches.
• start_time format: Must be ISO 8601 with Z: 2026-04-28T00:00:00Z
• API cost: $0.005 per tweet read. A 50-tweet fetch = $0.25.
• Rate cap: 2M reads/month on pay-per-use. Budget accordingly.
• Engagement filtering: Always fetch more tweets than needed since min_faves/min_retweets operators are not available on pay-per-use.

Verification

After fetching, confirm:

1. Returned tweets meet the engagement thresholds
2. No replies or retweets unless caller requested them
3. Tweets are within the lookback window
4. Author usernames and URLs are resolved correctly