Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.gdeltcloud.com/llms.txt

Use this file to discover all available pages before exploring further.

​
Overview

gdelt_system_prompt teaches agents how to use GDELT Cloud as an analyst product:
  • use Progressive Discovery wrappers
  • treat structured Events as the primary differentiator
  • use clustered Stories for narrative and article evidence
  • link Entities to Events and Stories
  • use macro-finance, prediction-market, and web-research tools as enrichment
  • avoid overusing repeated search_events calls when summaries, Stories, Entities, and enrichment tools are needed
  • start broad because generated Event coding and Story clustering are useful but imperfect
  • use default confidence_profile=precise Event/Story feeds unless exhaustive/raw recall is explicitly needed
  • answer in BLUF/inverted-pyramid order with a clear “So What” for analytical answers
  • place “Watch Next” before detailed evidence
  • use timeline tables for escalation, sequence, trend, or daily-change answers
  • render selected structured Events as gdelt-event blocks for GDELT Cloud inline cards
  • use clean Markdown and light emoji signposts where they improve Agent UI readability
  • label citations with source names/titles instead of showing bare full URLs
  • adapt prose depth to the user’s request
  • cite public GDELT Cloud URLs and source-article URLs

​
What The Prompt Emphasizes

AreaGuidance
Tool discoveryAlways use *_tool_list, inspect with *_tool_get, execute with *_tool_call.
GDELT Cloud anchorStart with GDELT Cloud structured Events, Stories, summaries, and Entities.
Tool balanceDo not tunnel on one tool; combine summaries, Events, Stories/articles, Entities, and enrichment as the task requires.
Filter disciplineStart broad, then narrow. Event and Story list/summary tools default to confidence_profile=precise for analyst-ready results; use confidence_profile=loose only when exhaustive/raw retrieval is needed. Heavy filter stacking can hide relevant records when real-time coding or clustering is imperfect.
EnrichmentUse macro finance, prediction markets, and web research frequently when they add context.
ProseQuick questions get concise answers; broad tasks get structured briefs.
Brief stylePut the most important takeaway first, then details. Include “So What” so implications are clear.
Watch NextPut watch items before detailed evidence so users see implications early.
TimelineUse Markdown tables for escalation, sequence, trend, or daily-change questions.
Event cardsUse fenced gdelt-event blocks for up to three key structured Events.
ReadabilityUse clean Markdown, short headings, bullets, bold labels, and light emoji signposts when appropriate.
CitationsCite GDELT Cloud Story/Entity URLs and source-article URLs using readable Markdown link labels, not bare URLs.
Historical disciplineRespect requested time windows and avoid hindsight leakage.

​
Agent Pattern

Scan: summarize_events / summarize_stories / search_stories
Zoom: search_events / get_event / get_story / get_story_articles / get_entity
Enrich: macro_finance_* / prediction_market_* / web_research_*
Answer: bottom line first, so what, watch next, timeline/signal, evidence, enrichment

​
Usage

messages = await mcp_client.get_prompt("gdelt-cloud", "gdelt_system_prompt")
system_text = "\n\n".join(str(msg.content) for msg in messages)
Pair the system prompt with the relevant task skill. Skills provide the domain-specific workflow for geopolitical, financial/sanctions, supply-chain, location-security, and general research tasks.

​
Related

Progressive Discovery

Learn the GDELT Cloud MCP wrapper flow.

Skills

Domain-specific analyst workflows.