Predict.fun Subgraphs

A suite of three subgraphs indexing Predict.fun — a prediction market protocol on BNB Chain (Polymarket fork) with $1.7B+ volume and novel yield-bearing mechanics via Venus Protocol.

Subgraphs

1. predictfun-orderbook

Indexes orderbook activity across all CTF and NegRisk exchanges.

Entities: Markets, Orderbooks, OrderFilled/Matched/Cancelled events, Fee tracking, Account stats, NegRisk markets, TradeData timeseries with hourly/daily aggregations

Contracts:

2. predictfun-positions

Indexes position lifecycle — splits, merges, redemptions, and open interest tracking.

Entities: Conditions, UserPositions, MarketOpenInterest, Split/Merge/Redemption events, NegRisk conversions, TransferSingle events

Contracts:

3. predictfun-yield

Indexes Predict.fun's novel yield-bearing mechanics — Venus Protocol integration, reward distributions, and UMA oracle resolution.

Entities: TokenMappings (underlying/vToken pairs), YieldClaims, VTokenMints, RewardRounds, OracleRequests/Proposals/Settlements

Contracts:

Architecture

predict.fun (BNB Chain)
├── predictfun-orderbook    ── Fills, matches, fees, market registration
├── predictfun-positions    ── Splits, merges, redemptions, open interest
└── predictfun-yield        ── Venus yield, reward claims, oracle resolution

All subgraphs share:

• Network: BSC (BNB Smart Chain)
• Collateral: USDT (18 decimals)
• Start Block: 64,817,753
• Spec Version: 1.3.0

Best Practices Applied

• Bytes! IDs everywhere (cheaper than String!)
• @entity(immutable: true) on all event logs
• @derivedFrom for reverse lookups (no redundant storage)
• No eth_calls (events only)
• indexerHints: prune: auto for storage efficiency
• nonFatalErrors feature enabled
• concatI32(logIndex) for unique event IDs

MCP Server

An MCP (Model Context Protocol) server that gives AI agents structured access to all three subgraphs.

Setup

1. Get a Graph API key from Subgraph Studio (docs)

2. Add to your Claude Code config (~/.claude/settings.json):

{
  "mcpServers": {
    "predictfun": {
      "command": "npx",
      "args": ["predictfun-mcp"],
      "env": {
        "GRAPH_API_KEY": "your-api-key-here"
      }
    }
  }
}

Subgraph IDs are built in. Queries go through The Graph Gateway and are billed to your API key.

OpenClaw / Remote Agents (SSE)

# Dual transport — stdio + SSE on port 3850
GRAPH_API_KEY=your-key npx predictfun-mcp --http

# SSE only (for remote/server deployments)
GRAPH_API_KEY=your-key npx predictfun-mcp --http-only

# Custom port
MCP_HTTP_PORT=4000 GRAPH_API_KEY=your-key npx predictfun-mcp --http

Point your agent at: http://localhost:3850/sse

Tools (14)

Data Tools

Meta-Tools (agent reasoning layer)

Meta-tools return structured JSON so agents can reason programmatically over trader behavior and market quality — not just raw volume and OI.

Prompts (9)

Pre-built workflows: platform_overview, analyze_trader, market_deep_dive, yield_analysis, whale_alert, market_scanner, custom_query_examples, trader_persona_analysis, market_quality_scan

Subgraph Development

cd predictfun-<subgraph>
npm install
npx graph codegen
npx graph build
npx graph deploy predictfun-<subgraph> --version-label v0.0.1

License

MIT