AI Agents Meet Prediction Markets
Prediction markets are one of the most information-dense environments on the internet. Polymarket alone processes hundreds of millions in volume across thousands of markets. Now, autonomous AI agents are entering these markets — analyzing data, placing bets, and managing portfolios without human intervention.
But there is a fundamental problem: how does an AI agent verify the track record of a counterparty before entering a position? In traditional finance, credit scores and audited track records provide this. In prediction markets, there is nothing.
The Problem: Pseudonymous Wallets, Unverifiable Claims
Polymarket wallets are pseudonymous. A wallet address like 0x7a3f... might have a stellar track record — or it might be a brand new address created yesterday. Anyone can claim a 70% win rate on social media. Anyone can screenshot a winning position while hiding the losses.
This creates critical gaps for AI agents operating in prediction markets:
- No verifiable track records — an agent cannot programmatically confirm a counterparty's historical accuracy
- No identity continuity — a wallet with a poor track record can simply be abandoned for a fresh address
- No standardized scoring — win rate alone is misleading without context (bet sizing, market difficulty, sample size)
- No machine-readable credentials — even if track data exists, there is no standard format for agents to consume it
- No revocation or expiry — a great 2024 track record says nothing about 2026 performance
Leaderboards on Polymarket exist, but they are platform-specific, not machine-readable, and they cannot be cryptographically verified by a third party. An AI agent evaluating a counterparty needs more than a screenshot — it needs a verifiable, signed attestation of prediction performance.
The Solution: MT Prediction
MolTrust's wallet-to-DID bridge solves this by linking on-chain prediction market activity to decentralized identifiers and issuing verifiable credentials that any agent can check.
What MT Prediction provides:
→ Wallet-to-DID binding — links Polymarket wallet addresses to MolTrust DIDs with on-chain proof
→ Automatic trade sync — pulls and indexes complete trade history from supported prediction markets
→ Composite prediction score — a single 0–100 number capturing accuracy, profitability, volume, and consistency
→ PredictionTrackCredential — a W3C Verifiable Credential encoding the score, signed by MolTrust (Ed25519)
→ Public leaderboard — ranked by prediction score, filterable by market category and time range
How the Prediction Score Works
A single win-rate number is misleading. A trader who bets on 99% favorites and wins 95% of the time is not impressive. A trader who consistently identifies mispriced markets and achieves 60% accuracy with positive ROI is. Our composite prediction score captures this nuance across five weighted dimensions:
Win Rate (30%)
Historical accuracy across all resolved markets. Weighted by market difficulty — correctly predicting a 50/50 market counts more than a 90/10 favorite. Brier score calibration is applied to penalize overconfidence.
ROI (25%)
Risk-adjusted returns relative to capital deployed. Measures not just whether the trader wins, but whether they generate positive expected value. A 55% win rate with 2:1 payoffs scores higher than 80% accuracy on -300 favorites.
Sample Size (20%)
Statistical significance of the track record. A wallet with 5 trades and 100% accuracy gets a lower score than one with 200 trades and 62% accuracy. We apply a Bayesian prior that converges toward the true score as sample size grows.
Volume (15%)
Skin in the game. Total capital deployed across markets. A trader risking real capital demonstrates conviction. This dimension prevents gaming through micro-bets and ensures the track record reflects meaningful economic activity.
Recency (10%)
Active participation and freshness of data. A stellar 2024 track record that went dormant in 2025 should be discounted. Exponential decay is applied with a 90-day half-life, rewarding consistent, recent activity.
The PredictionTrackCredential
Once a prediction score is calculated, it can be issued as a W3C Verifiable Credential — a machine-readable, cryptographically signed attestation that any agent or protocol can verify independently.
The credential encodes the complete score breakdown, the raw metrics, and the time period it covers. It expires after 90 days, ensuring scores stay current. Any agent can verify the signature against MolTrust's public DID without trusting a centralized API.
API Integration: 5 Endpoints
MT Prediction exposes five endpoints. All are free during Early Access. No API key required.
1. Link Wallet
Register a Polymarket wallet address and bind it to a MolTrust DID.
2. Get Prediction Score
Retrieve the composite prediction score for any linked wallet.
3. Issue PredictionTrackCredential
Issue a signed W3C Verifiable Credential for the wallet's prediction track record.
4. Verify Credential
Verify any PredictionTrackCredential — check signature, expiry, and score freshness.
5. Leaderboard
Public leaderboard of linked wallets ranked by prediction score.
Use Cases
Agent Counterparty Verification
An AI agent checks a counterparty's PredictionTrackCredential before entering a shared market position. Only engages with wallets scoring above 50.
Trading Desk Credibility
Quantitative trading desks verify wallet track records before following signals or entering copy-trade arrangements with prediction market operators.
DeFi Protocol Gating
DeFi protocols require a minimum prediction score to access premium market pools, oracle roles, or governance voting on prediction-related proposals.
Competitive Leaderboards
Prediction market communities run verifiable leaderboards. No more self-reported stats — every rank is backed by a signed credential with on-chain proof.
MCP Integration: 3 New Tools
MT Prediction is available as MCP tools for Claude Code and Claude Desktop. Add the moltrust-mcp-server to your MCP configuration and these tools become available directly in your AI workflow.
| Tool | Description | Parameters |
|---|---|---|
get_prediction_score |
Get the composite prediction score for a Polymarket wallet | wallet_address, platform |
verify_prediction_credential |
Verify a PredictionTrackCredential's signature, expiry, and score | vc (the credential JSON) |
get_prediction_leaderboard |
Retrieve the top prediction market participants ranked by score | limit, platform, category |
Pricing
All MT Prediction endpoints are free during Early Access. No API key. No signup. No rate limits beyond fair use.
When Early Access ends, MT Prediction will transition to x402 micropayments on Base — the same model used across MoltGuard. Pay per API call with USDC. No subscriptions. No invoices. Just cryptographic proof of payment in the HTTP header.
Why This Matters
Prediction markets are the closest thing to a ground truth engine on the internet. They aggregate information, price risk, and resolve disputes with cold, verifiable outcomes. But without verifiable identity and track records, the ecosystem cannot scale trust.
If you cannot verify who you are trading with, you cannot trust the market you are trading in.
MT Prediction brings the same trust infrastructure that MoltGuard provides for on-chain agents — and MT Shopping provides for agentic commerce — to the prediction market vertical. Verifiable credentials. Composite scores. On-chain anchoring. Open standards.
Get Started
- View the live leaderboard
- Read the full API reference
- Visit the MT Prediction landing page
- Install the MCP server from PyPI
- Learn about MoltGuard
Bridge Your Polymarket Wallet Today
Link your wallet, get your prediction score, and start building a verifiable track record. Free during Early Access.
Get Started →MT Prediction is built by MolTrust (CryptoKRI GmbH, Zurich) — trust infrastructure for the agent economy. Follow @MolTrust on X for updates.