x-social

X Social

Use xeng-mcp tools only. The host already supplies credentials and upstream routing — do not invent REST clients, env exports, or endpoint URLs.

When to use

• Topic clustering (themes, narratives, communities)
• Trend comparison across time windows
• Campaign / hashtag monitoring
• Market and sales signal reading (intent, complaints, competitors)

Rules

1. Call xeng_health once if connectivity is uncertain; otherwise start with xeng_search.
2. Prefer several focused xeng_search calls over one vague query.
3. Cite evidence with tweet_id and screen_name from returned rows only.
4. Treat pagination as a sample, not a full corpus or market share.
5. Never fabricate tweets, authors, or engagement. Never print secrets.
6. Do not shell out to HTTP, invent paths, or ask the user for a base URL — MCP tools are the API.

xeng_search

Required: q.

Optional: page, limit (max 100), offset, lang, screen_name, hashtag (no #), mention (no @), from_created_at, to_created_at, include_raw_json.

xeng_search:
  q: "<topic or keyword>"
  limit: 50
  hashtag: "<tag without #>"
  from_created_at: "<ISO lower bound>"
  to_created_at: "<ISO upper bound>"

Use response metadata.pagination when present for sample size; do not invent totals.

Procedures

Clustering

1. Seed search on the topic (limit 50–100); add synonym / slang / competitor queries as needed.
2. Group hits by theme (praise, complaint, meme, policy, spam, product, other).
3. Per cluster: label, sample size, representative quotes, recurring hashtags/mentions.
4. State that clusters are sample-based.

Trends

1. Fix q (and optional hashtag / mention).
2. Repeat xeng_search across sequential from_created_at / to_created_at windows.
3. Compare sample counts and engagement fields present on rows; note rising/falling themes and amplifiers only from data.
4. Empty window → report empty; do not extrapolate.

Campaign monitoring