Verg API
Public, agent-native access to curated convergence intelligence. No authentication required. Rate-limited to 60 requests/minute per IP.
Quick Start
Endpoints
/api/patternsReturns current convergence patterns with full provenance. Each pattern includes the CI score, independence verification, token bake cost, and links back to raw sources.
limit— max patterns (default 10)min_ci— minimum CI score filter (0-1)
{
"meta": { "total_patterns": 12, "generated_at": "..." },
"patterns": [
{
"id": "cp_e40f7a3c",
"label": "AI agents need human supervision",
"ci_score": 0.26,
"independence_score": 0.68,
"token_cost": { "raw_tokens": 18421, "curated_tokens": 723, "savings_percent": 96.1 },
"signal_quality": { "independence": "high", "platforms": [...] },
"sources": [...]
}
]
}/api/leadersThought leaders ranked by contribution score — not follower count. Each leader has originality, independence, centrality, and source depth metrics.
limit— max leaders (default 10)type— filter: architect, philosopher, amplifier, contrarian, bridgemin_score— minimum leader score
/api/openapi.jsonFull OpenAPI 3.0 specification. Machine-readable — drop it into Claude Code, Cursor, or any agent framework that supports OpenAPI tool discovery.
/api/feedbackSubmit feedback on a pattern or source. Agents and humans can both provide feedback to improve source quality scoring and pattern validation over time.
Key Concepts
CI Score (Convergence Intelligence)
Strength of convergence (0-1). Measures how many independent sources align and how closely their framing matches. 0.7+ is strong convergence.
Independence Score
Measured via social graph analysis (PageRank, Louvain communities). 0.7+ = strong cross-community convergence. Below 0.5 = potential echo chamber. Note: captures surface-level dependence (citations, affiliations) but not hidden common causes.
Token Bake
The processing cost of each insight. Raw source tokens in → curated structured artifact tokens out. Both numbers are measured from disk every run (chars/4 of the actual content), not estimated. Typical compression sits in the ~95-97% range — querying Verg returns the structured intelligence object instead of forcing the agent to process every contributing source itself. See /protocol#token-bake for the live measured numbers.
Leader Contribution
Scored on 4 dimensions: originality (novel ideas), independence (not echoing), centrality (pattern involvement), source depth (deep work vs shallow takes). NOT based on follower count.
Agent Integration
MCP Server
For Claude Code, Cursor, Windsurf, and other MCP-compatible tools, Verg offers a native MCP server with 5 tools: verg_patterns, verg_leaders, verg_emerging, verg_search, verg_predictions.
Discovery
- llms.txt — human-readable agent guide
- openapi.json — machine-readable tool spec
- whitepaper — protocol design