Skip to main content

​
Overview

GDELT Cloud Briefs are structured intelligence products generated by the hosted Brief agent and rendered as shareable Web Brief reports. The Brief agent does not improvise its own report format. It preloads:
  • gdelt_brief_system_prompt from the MCP server
  • the shared Brief tradecraft skill from MCP resources
  • the selected Brief production skill from MCP resources
  • the matching briefs.v1 structured content schema from the shared contracts package
The model populates only the analysis/content fields. Runtime software appends deterministic metadata, citations, appendix/tool diagnostics, run IDs, share/export state, and persistence fields.

​
Brief Types

Brief typeSkill purposeStrong public pattern libraries
Daily Operational Risk BriefConcise morning intelligence: what changed, why it matters, and what to watch.CIA President’s Daily Brief, ISW daily campaign assessments
Crisis SITREP / Flash UpdateOperational update for an active incident, disruption, attack, protest, disaster, or outage.UN OCHA SITREPs, UKMTO/JMIC maritime advisories
Indications & Warning Watch BriefForward-looking warning built from observable indicators and tripwires.CrisisWatch, DHS NTAS, CISA advisories
Country, Sector & Supplier Exposure MonitorRecurring exposure monitor for countries, sectors, suppliers, assets, routes, ports, and portfolios.Control Risks RiskMap, IRC Watchlist, IMB piracy reports

​
Evidence Workflow

Every Brief skill follows the same evidence loop: 
Structured event signal
→ baseline comparison
→ Story cluster evidence
→ corroboration
→ operational relevance
→ key judgment
→ confidence / caveat
→ watch indicator
→ finished Brief content
GDELT Cloud Events provide the structured signal. GDELT Cloud Stories provide the narrative and source/article layer. External MCP categories enrich the analysis when relevant:
External categoryUse in Briefs
web_research_*Official statements, emergency guidance, port/maritime advisories, procurement pages, source extraction, and corroboration.
macro_finance_*Commodities, FX, rates, equities, shipping, energy, sovereign-risk, and sector-channel exposure.
prediction_market_*Market-implied probability context only when the contract maps cleanly to the scenario. Never use markets as ground truth.
Open-web AIS, ADS-B, ports, USAspending, SAM.gov, Open Contracting, Global Fishing Watch, OpenSky, PortWatch, and similar sources are treated as corroboration targets through web_research_* until GDELT Cloud exposes normalized native connectors for them. Briefs should preserve visual evidence from tool payloads when available. Event and Story cards can carry top-article image_url, article domain_avatar_url, and linked Entity image_url / avatar_url. Entity tools can carry Wikipedia thumbnail metadata. These fields should improve reader orientation and source recognition, not substitute for analytic evidence. For time-series panels, Brief agents should collect real history:
  • GDELT Cloud summaries: call summarize_events or summarize_stories with group_by=date over a recent window such as 14 or 30 days, then compare the current bucket against baseline or prior buckets.
  • Macro finance: use provider-supported historical or series tools; normalized outputs expose series, latest_value, and latest_date when history is available.
  • Prediction markets: use include_history=true where available; normalized outputs may expose probability_series.
Do not render or describe two-point placeholder trends as meaningful analysis. If a tool only returns a current value, say that the series is unavailable and use the value as context, not proof.

​
Separation Of Concerns

LayerOwns
MCPPrompts, tools, resources, task skills, Brief skills, code/reference resources.
Shared contractsbriefs.v1 input schemas, content schemas, artifact schema, registry.
Brief agentFetches MCP prompt/skills, uses MCP tools, applies the selected structured response schema, returns Brief content.
GDELT Cloud appValidates inputs, starts one-off or scheduled Brief runs, persists artifacts, renders Web Brief reports, and manages sharing.
Brief storeStores Briefs, schedules, batches, runs, share keys, plan limits, and usage logs.
Agents and frontend code should know only stable MCP identifiers such as prompt names, skill URIs, and tool category tags. The details of GDELT Cloud tool schemas and analysis tradecraft live in MCP.

​
Structured Output

Brief content schemas contain only agent-populated analytical content:
  • BLUF / key takeaway
  • key judgments with confidence
  • material changes versus baseline
  • evidence-backed developments
  • affected areas, actors, timelines, scorecards, or indicators depending on Brief type
  • outlook
  • watch indicators
  • recommended actions
  • caveats and uncertainty
Software appends:
  • schema/artifact version
  • run and thread IDs
  • citations and source indexes
  • appendix/tool diagnostics
  • share and export state
  • owner/public visibility controls

​
Scheduling And Plans

Brief schedules use the same knobs as Alerts:
KnobValues
Enabledon/off
Frequencyhourly, daily, weekly
Allowanceplan-level max_briefs; null means unlimited
Minimum frequencyplan-level min_brief_frequency; for example daily allows daily and weekly but blocks hourly
Scheduled Briefs are queued and processed by the same internal cron pattern as Alerts, with admin testing endpoints and an admin console.

​
Query Units

Direct MCP access and hosted agent access are metered differently:
CallerUsage sourceQU per MCP data tool call
User’s own MCP client, ChatGPT connector, Claude connector, or custom LangChain agentmcp1 QU
GDELT Cloud Agent UIagent5 QU
GDELT Cloud hosted Brief agent, including scheduled Briefsbrief0 QU; metered in BU
The Agent multiplier applies to GDELT Cloud data tools and external MCP categories. Briefs are metered separately in Brief Units, so their tool calls are recorded for reporting but do not debit QU.

​
Related

System Prompt

MCP prompts used by general and Brief agents.

Progressive Discovery

How agents discover and call MCP tool categories.