Imported from ChainAware/behavioral-prediction-mcp (SKILL.md). Install upstream with npx skills add ChainAware/behavioral-prediction-mcp. Copyright stays with the author (MIT).

ChainAware Behavioral Prediction MCP

What This Skill Does

The ChainAware Behavioral Prediction MCP connects any AI agent to a continuously updated Web3 behavioral intelligence layer: 14M+ wallet profiles across 8 blockchains, built from 1.3 billion+ predictive data points. It delivers fourteen capabilities via a single MCP endpoint:

1. Fraud Detection — predict fraudulent wallet behavior before it happens (~98% accuracy on ETH)
2. Batch Fraud Detection — async batch job for fraud screening large wallet lists; returns job_id + signature
3. Behavioral Analysis — profile wallet intent, risk tolerance, experience, and next likely actions
4. Batch Behavioral Analysis — async batch job for behavioral profiling large wallet lists; returns job_id + signature
5. Batch Job Status — poll the progress of any running batch job
6. Batch Job Results — retrieve full per-wallet results from a completed or partial batch job
7. Rug Pull Detection — forecast whether a smart contract or liquidity pool will rug pull
8. Credit Score — crypto credit/trust score (1–9) combining fraud probability and social graph analysis
9. Token Rank List — rank tokens by holder community strength across chains and categories
10. Token Rank Single — deep-dive into a single token's community quality and top holders
11. Token Audit — deep multi-module smart contract audit via async get-or-create pipeline; 8 modules (ownership, liquidity, supply, honeypot, reentrancy, permissions, shadow functions, drainability); aggregate risk score 0–100; supports eth, bsc, base, arbitrum, avalanche, optimism, polygon (lowercase)
12. Token Audit Results — poll or retrieve completed token audit; full module breakdown when audit_status == "complete"; never present data while still queued or running
13. Agent Trust Score List — list ERC-8004 registered AI agents with on-chain trust scores (0–1000); paginated and sortable; use to discover agents or resolve agent_id
14. Agent Trust Score Single — full trust profile for a single ERC-8004 agent by agent_id + chain_id; includes trust score, tier, wallet_verified, owner_address; hard warning if wallet_verified == false or error is non-null

Unlike forensic blockchain tools that describe the past, this MCP is predictive — it tells your agent what is about to happen.

MCP Server URL: https://prediction.mcp.chainaware.ai/sse
GitHub: https://github.com/ChainAware/behavioral-prediction-mcp
Website: https://chainaware.ai
Pricing / API Key: https://chainaware.ai/pricing · x402 payment supported
Twitter: https://x.com/ChainAware/
LinkedIn: https://www.linkedin.com/company/chainaware
Blog: https://chainaware.ai/blog
Learn: https://chainaware.ai/learn
Examples: https://github.com/ChainAware/examples
Fraud Detection Accuracy: 98% backtesting verified
Rug Pull Detection Accuracy: 90.1% backtesting verified
Recognition: CB Insights Fraud Prevention Market Map (2026) · BNB Chain AI Landscape (2025) · BNB Chain Kickstart (2025) · Google Cloud $250k Grant (2025) · AWS Fintech Accelerator (2024) · Safary Club Web3 Growth Landscape (2024)

Capabilities

• Fraud Detection — predict fraudulent wallet behavior before it happens (~98% accuracy on ETH)
• Batch Fraud Detection — async batch fraud screening for large wallet lists; fire-and-fetch pattern via predictive_fraud_batch → check_job_status → get_job_results
• Behavioral Analysis — profile wallet intent, risk tolerance, experience, and next likely actions across DeFi, NFT, and trading segments
• Batch Behavioral Analysis — async batch behavioral profiling for large wallet lists; same fire-and-fetch pattern via predictive_behaviour_batch → check_job_status → get_job_results
• Rug Pull Detection — forecast whether a smart contract or liquidity pool will rug pull
• Credit Score — crypto credit/trust score (1–9) combining fraud probability and social graph analysis for DeFi lending decisions
• Token Rank List — rank tokens by holder community strength across ETH, BNB, BASE, and Solana
• Token Rank Single — deep-dive into a specific token's community quality and top holders

When to Use This Skill

• User asks about wallet safety, fraud risk, or suspicious activity
• User wants to screen a wallet, contract, or LP before interacting with it
• User needs AML/compliance checks on a blockchain address
• User wants behavioral profiling or DeFi personalization for a wallet
• User asks about token quality, community strength, or holder analysis
• User is building a DeFi platform, AI agent, launchpad, or compliance tool
• User wants to integrate the ChainAware MCP into their codebase

When NOT to Use This Skill

• User asks about general blockchain data (balances, transaction history) → use a block explorer
• User wants real-time price data or market cap → use a market data API
• User wants to analyze smart contract code for bugs → use a code auditing tool
• For complex behavioural analysis (deep wallet profiling including fraud signals) → escalate to chainaware-wallet-auditor subagent
• For batch screening of many wallets → use batch MCP tools (predictive_fraud_batch / predictive_behaviour_batch) for large lists, or chainaware-fraud-detector subagent for small lists
• For marketing personalization → use chainaware-wallet-marketer subagent

Supported Blockchains

Step-by-Step Workflow

For wallet fraud screening

1. Confirm inputs — wallet address and network. If network is missing, ask.
2. Call predictive_fraud with the wallet address and network.
3. Interpret probabilityFraud using the threshold table below.
4. Scan forensic_details for negative flags (mixer use, sanctioned entities, darknet, etc.).
5. Report status, score, and any forensic flags in plain language.

For behavioral profiling / personalization

1. Confirm inputs — wallet address and network.
2. Call predictive_behaviour with the wallet address and network.
3. Extract key signals: intention.Value (Prob_Trade/Stake/Bridge/NFT_Buy), experience.Value, categories, recommendation.
4. Classify the wallet by dominant category and intent signal.
5. Generate personalized recommendations or next-best-action based on the profile.

For rug pull / contract safety checks

1. Confirm inputs — smart contract or LP address and network.
2. Optionally call predictive_fraud on the deployer address first for extra signal.
3. Call predictive_rug_pull with the contract address.
4. Interpret probabilityFraud and scan forensic_details for liquidity and contract risk flags.
5. Apply the Deployer Risk Amplifier: if deployer fraud score ≥ 0.5, escalate overall risk one level.
6. Report verdict with supporting forensic evidence.

For token ranking / discovery

1. Identify the request — list of tokens or single token deep-dive?
2. For lists: call token_rank_list with appropriate category, network, sort_by: communityRank, sort_order: DESC.
3. For single tokens: call token_rank_single with contract_address and network.
4. Report communityRank, normalizedRank, totalHolders, and top holder profiles.

For batch fraud or behavioural screening (large wallet lists)

1. Schedule — call predictive_fraud_batch or predictive_behaviour_batch with the wallet list and network.
2. Store both job_id and signature from the response — required for all follow-up calls.
3. Poll — call check_job_status with job_id + signature until status is completed or partial.
   • Status pending or processing → wait and retry.
   • Status partial → some wallets failed but results are available for completed items.
4. Retrieve — call get_job_results with job_id + signature to fetch the full per-wallet data.
5. Process — the data array returns the same schema as single-wallet predictive_behaviour / predictive_fraud results.

Never call get_job_results while status is still pending or processing. The job expires_at timestamp in the status response indicates how long results are retained.

For full due diligence (multi-tool)

1. Call predictive_fraud → get fraud score and forensic flags
2. Call predictive_behaviour → get behavioral profile and intent
3. Call predictive_rug_pull (if a contract address) → get contract risk
4. Synthesize all three into a unified verdict with risk level and recommendation

For complex due diligence workflows, escalate to the chainaware-wallet-auditor subagent.

Risk Score Thresholds

Available Tools

1. predictive_fraud — Fraud Detection

Forecasts the probability that a wallet will engage in fraudulent activity. Includes AML checks. Use when a user wants to screen a wallet before interacting with it.

Inputs:

• apiKey (string, required) — ChainAware API key
• network (string, required) — e.g. ETH, BNB, BASE
• walletAddress (string, required) — the wallet to evaluate

Key output fields:

• status — "Fraud", "Not Fraud", or "New Address"
• probabilityFraud — decimal 0.00–1.00
• forensic_details — deep on-chain breakdown

Example prompts that trigger this tool:

• "Is it safe to interact with 0xABC... on Ethereum?"
• "What is the fraud risk of this BNB wallet?"
• "Run an AML check on this address."
• "Screen this wallet before onboarding."
• "Is this address on any sanctions list?"
• "Pre-screen this user's wallet for compliance."

2. predictive_behaviour — Behavioral Analysis & Personalization

Profiles a wallet's on-chain history and predicts its next actions.

Inputs:

• apiKey (string, required)
• network (string, required)
• walletAddress (string, required)

Key output fields:

• intention — predicted next actions (Prob_Trade, Prob_Stake, Prob_Bridge, Prob_NFT_Buy — High/Medium/Low)
• recommendation — personalized action suggestions
• categories — behavioral segments (DeFi Lender, NFT Trader, Bridge User, etc.)
• riskProfile — risk tolerance and balance age breakdown
• experience — experience score 0–10 (beginner → expert)
• protocols — which protocols this wallet uses (Aave, Uniswap, GMX, etc.)

Example prompts that trigger this tool:

• "What will this wallet do next?"
• "Is this user a DeFi lender or an NFT trader?"
• "Recommend the best yield strategy for this address."
• "What's the experience level of this wallet?"
• "Personalize my DeFi agent's response for this user."
• "Segment this wallet for my marketing campaign."

3. predictive_rug_pull — Rug Pull Detection

Forecasts whether a smart contract or liquidity pool is likely to execute a rug pull.

Inputs:

• apiKey (string, required)
• network (string, required)
• walletAddress (string, required) — smart contract or LP address

Key output fields:

• status — "Fraud" or "Not Fraud"
• probabilityFraud — decimal 0.00–1.00
• forensic_details — on-chain metrics behind the score

Example prompts that trigger this tool:

• "Will this new DeFi pool rug pull if I stake my assets?"
• "Is this smart contract safe?"
• "Check if this launchpad project is legitimate."
• "Monitor this LP position for rug pull risk."
• "Is this contract deployer trustworthy?"

4. credit_score — Crypto Credit Score

Calculates a credit/trust score (1–9) for a wallet by combining fraud probability with social graph analysis. Designed for DeFi lending and any use case needing a fast single-number creditworthiness signal.

Inputs:

• apiKey (string, required)
• network (string, required) — ETH
• walletAddress (string, required) — the wallet to score

Key output fields:

• creditData.riskRating — integer 1–9 (1 = highest risk, 9 = highest trust)
• creditData.walletAddress — echoed wallet address

Example prompts that trigger this tool:

• "What is the credit score for 0xABC...?"
• "Is this wallet a reliable borrower?"
• "Calculate credit score for this address on ETH."
• "Rate this wallet's creditworthiness."
• "Trust score for lending — 0xDEF... on BNB."

5. token_rank_list — Token Ranking by Holder Strength

Ranks tokens by the quality and strength of their holder community.

Inputs:

• limit (string, required) — items per page
• offset (string, required) — page number
• network (string, required) — ETH, BNB, BASE, SOLANA
• sort_by (string, required) — e.g. communityRank
• sort_order (string, required) — ASC or DESC
• category (string, required) — AI Token, RWA Token, DeFi Token, DeFAI Token, DePIN Token
• contract_name (string, required) — token name search (empty string for no filter)

Key output fields:

• data.total — total matching tokens
• data.contracts[] — array with contractAddress, contractName, ticker, chain, category, communityRank, normalizedRank, totalHolders

Example prompts that trigger this tool:

• "What are the top AI tokens on Ethereum?"
• "Rank DeFi tokens on BNB by community strength."
• "Which RWA tokens have the strongest holder base on BASE?"
• "Show me the top 10 tokens by community rank on ETH."
• "Compare DePIN tokens across Solana and Ethereum."

6. token_rank_single — Single Token Rank & Top Holders

Returns the rank and top holders for a specific token by contract address.

Inputs:

• contract_address (string, required) — token contract or mint address
• network (string, required) — ETH, BNB, BASE, SOLANA

Key output fields:

• data.contract — token details including communityRank, normalizedRank, totalHolders
• data.topHolders[] — holder wallet addresses with balance, walletAgeInDays, transactionsNumber, totalPoints, globalRank

Example prompts that trigger this tool:

• "What is the token rank for USDT on Ethereum?"
• "Who are the top holders of 0xdAC17F... on ETH?"
• "How strong is the holder base of this contract on BNB?"
• "Show me the best holders of this Solana token."

7. predictive_fraud_batch — Batch Fraud Detection (Schedule)

Schedules an async fraud detection job for a list of wallet addresses. Returns a job handle immediately — results are fetched later via check_job_status + get_job_results.

Inputs:

• apiKey (string, required) — ChainAware API key
• network (string, required) — ETH, BNB, POLYGON, TON, BASE, TRON, HAQQ
• addresses (array[objects], required) — list of wallet address objects to evaluate

Key output fields:

• job_id — unique job identifier (store this)
• signature — access token for follow-up calls (store this)
• status — always "pending" on schedule response
• total_items — number of wallets submitted

Example prompts that trigger this tool:

• "Run fraud screening on this list of 500 wallets on ETH."
• "Batch AML check for these addresses on BNB."
• "Screen all wallets from this CSV for fraud on BASE."

8. predictive_behaviour_batch — Batch Behavioral Analysis (Schedule)

Schedules an async behavioral analysis job for a list of wallet addresses. Same fire-and-fetch pattern as predictive_fraud_batch.

Inputs:

• apiKey (string, required) — ChainAware API key
• network (string, required) — ETH, BNB, BASE, HAQQ, SOLANA
• addresses (array[objects], required) — list of wallet address objects to evaluate

Key output fields:

• job_id — unique job identifier (store this)
• signature — access token for follow-up calls (store this)
• status — always "pending" on schedule response
• total_items — number of wallets submitted

Example prompts that trigger this tool:

• "Profile all wallets in this list on ETH — intent, experience, risk."
• "Batch behavioural analysis for these 200 addresses on BASE."
• "Run segment analysis across all wallets in this Solana list."

9. check_job_status — Batch Job Progress

Checks the progress of a scheduled batch job. Returns counts only — no wallet data. Call this after scheduling a batch job and before fetching results.

Inputs:

• job_id (string, required) — from predictive_fraud_batch or predictive_behaviour_batch
• signature (string, required) — from the same schedule call

Key output fields:

• status — "pending" | "processing" | "partial" | "completed"
• completed_items, failed_items, pending_items — progress counts
• expires_at — when results will be purged

10. get_job_results — Batch Job Results

Retrieves the full per-wallet results from a completed or partial batch job. Returns the same rich schema as the single-wallet tools. Only call when check_job_status shows completed or partial.

Inputs:

• job_id (string, required) — from the schedule call
• signature (string, required) — from the same schedule call

Key output fields:

• data[] — array of per-wallet results; each entry mirrors predictive_behaviour / predictive_fraud output schema

Example prompts that trigger this tool:

• "Get the results for job 0fc5897a on ETH."
• "Fetch the completed batch fraud results."
• "Retrieve wallet profiles from the batch I scheduled earlier."

Validation Checkpoints

Input Validation

• ✅ Wallet address provided and non-empty
• ✅ Network specified and supported for the tool being called (check table above)
• ✅ CHAINAWARE_API_KEY environment variable is set (not required for check_job_status / get_job_results)
• ✅ For token_rank_list: limit, offset, sort_by, sort_order, and category all provided
• ✅ For token_rank_single: both contract_address and network provided
• ✅ For batch tools: both job_id and signature stored from the schedule response before calling follow-up tools
• ⚠️ If network is missing, ask the user before proceeding
• ⚠️ If network is not supported for the requested tool, inform the user and suggest an alternative
• ⚠️ Never call get_job_results while job status is pending or processing

Output Validation

• ✅ probabilityFraud is present and in range 0.00–1.00
• ✅ Risk threshold label applied correctly (see table above)
• ✅ Forensic flags surfaced in plain language, not raw JSON
• ✅ Every recommendation cites the specific signal that drove it
• ✅ Network limitations clearly stated when a tool doesn't support the requested chain
• ✅ For behavioral profiles: at least intention, experience, and categories included in response
• ✅ For batch jobs: always report job_id to the user after scheduling; they may need it to check status manually

Example Output

Fraud Check — 0xABC... on ETH

🔮 FRAUD ASSESSMENT
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Wallet:  0xABC...
Network: ETH
Status:  🟡 MEDIUM RISK

Fraud Probability: 0.34
Risk Level: Medium — proceed with caution

Forensic Highlights:
  • 3 transactions flagged as suspicious
  • No mixer/tumbler activity detected
  • No sanctioned entity connections
  • Wallet age: 187 days

Recommendation: Monitor this wallet. Not safe for large-value
interactions without additional verification.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Behavioral Profile — 0xDEF... on BASE

🧠 BEHAVIORAL PROFILE
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Wallet:  0xDEF...
Network: BASE

Experience:   7.2/10 — Experienced
Segment:      DeFi Lender, Bridge User
Risk Profile: Balanced

Intent Signals:
  Trade:    High
  Stake:    Medium
  Bridge:   High
  NFT Buy:  Low

Protocols Used: Aave, Uniswap, Across Bridge

Recommendation:
  → Promote yield optimization vaults
  → Highlight cross-chain bridging incentives
  → Skip NFT-focused messaging
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Requirements

• API Key — a CHAINAWARE_API_KEY environment variable is required for subscription-based access. Obtain one at https://chainaware.ai/pricing
• x402 payments — the MCP server supports the x402 protocol for pay-per-call access without a subscription; compatible clients settle payment automatically on HTTP 402 responses
• MCP-compatible host — Claude Code, Cursor, Claude Desktop, ChatGPT Connectors, or any MCP client that supports SSE transport
• Network awareness — different tools support different blockchains; see the Supported Blockchains table above
• No local installation — the MCP server runs remotely at https://prediction.mcp.chainaware.ai/sse; no packages to install

Integration Setup

Claude Code (CLI)

claude mcp add --transport sse chainaware-behavioural-prediction-mcp-server \
  https://prediction.mcp.chainaware.ai/sse \
  --header "X-API-Key: your-key-here"

📚 Docs: https://code.claude.com/docs/en/mcp

Claude Web / Claude Desktop

1. Go to Settings → Integrations → Add integration
2. Name: ChainAware Behavioural Prediction MCP Server
3. URL: https://prediction.mcp.chainaware.ai/sse?apiKey=your-key-here

📚 Docs: https://platform.claude.com/docs/en/agents-and-tools/remote-mcp-servers

Cursor (mcp.json)

{
  "mcpServers": {
    "chainaware-behavioural-prediction-mcp-server": {
      "url": "https://prediction.mcp.chainaware.ai/sse",
      "transport": "sse",
      "headers": {
        "X-API-Key": "your-key-here"
      }
    }
  }
}

📚 Docs: https://cursor.com/docs/context/mcp

ChatGPT Connectors

1. Open ChatGPT Settings → Apps / Connectors → Add Connector
2. Name: ChainAware Behavioural Prediction MCP Server
3. URL: https://prediction.mcp.chainaware.ai/sse?apiKey=your-key-here

Node.js

import { MCPClient } from "mcp-client";
const client = new MCPClient("https://prediction.mcp.chainaware.ai/");

const fraud = await client.call("predictive_fraud", {
  apiKey: process.env.CHAINAWARE_API_KEY,
  network: "ETH",
  walletAddress: "0xYourWalletAddress"
});

const topTokens = await client.call("token_rank_list", {
  limit: "10", offset: "0", network: "ETH",
  sort_by: "communityRank", sort_order: "DESC",
  category: "AI Token", contract_name: ""
});

Python

from mcp_client import MCPClient
import os

client = MCPClient("https://prediction.mcp.chainaware.ai/")
result = client.call("predictive_fraud", {
    "apiKey": os.environ["CHAINAWARE_API_KEY"],
    "network": "ETH",
    "walletAddress": "0xYourWalletAddress"
})

Real-World Use Cases

DeFi Platforms

• Risk-adjusted lending — use fraud scores and behavioral profiles to set collateral requirements and interest rates per borrower
• Liquidity management — use intent signals to pre-position reserves and prevent pool drain
• Yield routing — identify wallets with high yield-seeking intent and route them to optimal vaults

AI Agent Personalization

• Give your agent a real-time behavioral profile of each wallet it talks to
• Segment users automatically into DeFi Lender, NFT Trader, Bridge User, New Wallet, etc.

Fraud & Compliance

• Screen wallets at the point of entry to your Dapp — before any transaction takes place
• Run AML monitoring across all active wallets
• Detect rug pull contracts at launchpad listing stage

NFT & GameFi

• Personalize in-game economies based on a player wallet's on-chain history
• Filter bot wallets and wash traders from NFT drops using fraud scores

Tips for Success

1. Always specify the network — many tools behave differently across chains
2. Run fraud check first — before any behavioral profiling, gate on fraud score
3. Combine tools for full due diligence — fraud + behaviour + rug pull together give a complete picture
4. Use the Deployer Risk Amplifier — a clean contract from a fraudulent deployer is still high risk
5. For small lists (< 5 wallets) — use subagents like chainaware-fraud-detector or chainaware-airdrop-screener which call single-wallet tools in a loop
6. For 5+ wallets — use batch tools directly: predictive_fraud_batch or predictive_behaviour_batch → check_job_status → get_job_results
7. Always store job_id + signature — both are required for every follow-up batch call; losing either means you cannot retrieve results
8. Surface forensic flags in plain language — never return raw JSON to end users

Related Subagents (Claude Code)

These subagents in .claude/agents/ provide specialized autonomous execution:

Background Reading

Data & Privacy

What data leaves your environment

Every tool call transmits the following to https://prediction.mcp.chainaware.ai/sse:

What is NOT sent: names, emails, IP addresses, private keys, raw transaction history, or any off-chain personal data.

API key handling

CHAINAWARE_API_KEY is read from the environment and passed as the apiKey parameter in each tool call. It is never included in output, never written to disk, and never logged by this skill. Treat it as a secret and rotate it regularly.

The MCP server also supports x402 payments — if your client supports the x402 protocol, payment is settled automatically on HTTP 402 responses with no API key required.

Integration-specific privacy notes

• Claude Code / Cursor: key passed via X-API-Key header — does not appear in URLs or logs
• Claude Web / ChatGPT: key must be appended to the SSE URL (?apiKey=...) — these platforms do not support custom SSE headers. Be aware the key will appear in your browser's network tab. Use a restricted-scope key for these integrations.

Operator responsibilities

Wallet addresses are pseudonymous identifiers. Whether they constitute personal data in your jurisdiction depends on your regulatory context (e.g. GDPR, MiCA). Operators processing wallets linked to identified users should perform their own data protection assessment.

Privacy policy: https://chainaware.ai/privacy

Security Notes

• Never hard-code API keys in public repositories
• The server uses SSE (Server-Sent Events) for streaming responses
• Rate limits apply depending on your subscription tier

Error Reference

Access & Pricing

Two access methods are supported:

• API key — subscribe at https://chainaware.ai/pricing and pass the key via CHAINAWARE_API_KEY env var
• x402 payments — pay-per-call via the x402 protocol; no subscription required, compatible clients handle payment automatically on HTTP 402 responses