Prediction MCP¶

The ChainAware Behavioral Prediction MCP is an open-source Model Context Protocol server that gives AI agents real-time predictive intelligence over blockchain wallets and smart contracts - without writing API integration code. It is the trust and fraud intelligence layer for agentic AI systems: fraud scoring, AML screening, Know Your Agent (KYA) verification, and rug pull detection, all consumable by any agent in a single tool call.

Any MCP-compatible AI environment - Claude Code, Claude Desktop, Cursor, or a custom multi-agent pipeline - can connect to the MCP server and immediately gain access to fourteen prediction tools backed by 23M+ wallet profiles across 8 blockchains.

As AI agents become autonomous economic actors - paying for services, subscribing, transacting 24/7 - the fraud prevention and identity systems built for humans don't carry over. The agentic economy needs its own native trust infrastructure. ChainAware's MCP server is that layer: a single tool call away from a real-time answer on whether a wallet, token, or pool is safe to interact with.

14
MCP Tools

34
Subagents

8
Blockchains

23M+
Wallet Profiles

MCP Endpoint: https://prediction.mcp.chainaware.ai/sse

GitHub: github.com/ChainAware/behavioral-prediction-mcp

How It Fits Your Agent's Decision Loop¶

01 - Agent proposes action
Your agent decides to swap, lend, pay, or counter-party with an on-chain address.

02 - MCP tool call
Agent calls get_trust_score / audit_token / check_rugpull_risk over SSE - no custom integration required.

03 - Structured risk object returned
Score, flags, and evidence return in milliseconds, sourced from live block scans.

04 - Agent acts or aborts
Execute, downsize, or decline the transaction autonomously - with the reasoning logged.

MCP Tools¶

Tool	What It Does	Networks
`predictive_fraud`	Fraud probability + AML forensics for a wallet address	ETH, BNB, POLYGON, TON, BASE, TRON, HAQQ
`predictive_fraud_batch`	Batch fraud detection for a list of wallets - schedules a job, returns `job_id` immediately	ETH, BNB, POLYGON, TON, BASE, TRON, HAQQ
`predictive_behaviour`	Wallet segmentation, intent prediction, experience scoring, personalisation recommendations	ETH, BNB, BASE, HAQQ, SOLANA
`predictive_behaviour_batch`	Batch behavioural profiling for a list of wallets - schedules a job, returns `job_id` immediately	ETH, BNB, BASE, HAQQ, SOLANA
`predictive_rug_pull`	Rug pull risk score for smart contracts and liquidity pools	ETH, BNB, BASE, HAQQ
`token_rank_list`	Ranked list of tokens by holder community strength	ETH, BNB, BASE, SOLANA
`token_rank_single`	Community rank + top holders for a specific token contract	ETH, BNB, BASE, SOLANA
`credit_score`	AI-driven crypto trust score (1-9) combining behavioural analysis, fraud scoring, and social graph signals	ETH
`run_token_audit`	Deep smart contract audit - ownership, liquidity, supply, transfer, approve, permit, reentrancy, and honeypot analysis	eth, bsc, base, arbitrum, avalanche, optimism, polygon
`get_token_audit_result`	Retrieve token audit results - poll this when `run_token_audit` returns `status: "queued"`	eth, bsc, base, arbitrum, avalanche, optimism, polygon
`agents_trust_score_list`	Paginated list of ERC-8004 AI agents with 0-1000 trust scores	ETH, BNB, Base, AVAX, Mantle
`agents_trust_score_single`	Full trust profile for a single ERC-8004 registered AI agent	ETH, BNB, Base, AVAX, Mantle

Agent Trust Score index (live): 377,608 agents indexed - average score 420, 106,254 flagged high-risk (0-200 tier). Behavior-based scoring derived from on-chain history, counterparty risk, and fraud probability - not peer voting, which can be Sybil-farmed with throwaway wallets.
| check_job_status | Check progress of a batch job - returns completed/failed/pending counts | - |
| get_job_results | Retrieve results of a completed or partial batch job | - |

Quick Setup¶

Free to test - no signup required. Point any MCP client at the SSE endpoint and start calling tools immediately. For production use, get an API key at chainaware.ai/support (also visible at chainaware.ai/profile).

Step 1 - Connect (Claude Code)

claude mcp add --transport sse chainaware-behavioral-prediction \
  https://prediction.mcp.chainaware.ai/sse \
  --header "X-API-Key: YOUR_API_KEY"

Step 1 - Connect (Claude Desktop, LangChain, custom stack)

{
  "mcpServers": {
    "chainaware": {
      "url": "https://prediction.mcp.chainaware.ai/sse"
    }
  }
}

Step 2 - Start using the tools

The prediction tools are now available in any MCP-compatible session. Optionally install the 34 pre-built subagents:

git clone https://github.com/ChainAware/behavioral-prediction-mcp.git
cp behavioral-prediction-mcp/.claude/agents/*.md .claude/agents/

Examples: github.com/ChainAware/examples - ready-to-run agent examples for fraud gating, token screening, and onboarding workflows.

Full setup instructions for all MCP clients: Setup Guide.

What You Can Build¶

The tools are composable - use them individually or combine them into more powerful workflows.

Fraud gate at wallet-connect
Call predictive_fraud at the moment a wallet connects to your dapp. Block wallets above your risk threshold before they interact with your protocol - in under 100ms.

Personalised onboarding
Call predictive_behaviour after connection to get experience level, behavioral categories, and personalised recommendations. Route DeFi veterans directly to advanced features; show newcomers a guided flow.

Rug pull screening
Call predictive_rug_pull before listing a token or pool on your DEX. Block high-risk contracts from appearing in your UI automatically.

Token discovery and due diligence
Call token_rank_list to find the strongest tokens in a category by holder community quality. Follow up with token_rank_single to deep-dive into a specific token's top holders.

Credit scoring for lending
Call credit_score to get a 1-9 trust rating for a borrower wallet. Combine with predictive_fraud for a complete lending risk picture before originating undercollateralised loans.

Full wallet intelligence pipeline
Chain the wallet tools together: predictive_fraud for risk, credit_score for lending trust, predictive_behaviour for profile - then apply the result to onboarding, credit, or personalisation decisions.

The 34 Subagents¶

The GitHub repository includes 34 pre-built Claude Code subagents that wrap these tools with focused prompts and decision logic - ready-made agents for fraud detection, AML scoring, compliance screening, onboarding routing, whale detection, credit assessment, GameFi screening, DAO governance, token contract auditing, and ERC-8004 agent trust screening.

See all 34 subagents →

Authentication¶

Every tool call requires an apiKey parameter. Set it as an environment variable rather than hardcoding it:

export CHAINAWARE_API_KEY="your-api-key-here"

Reference it in code as process.env.CHAINAWARE_API_KEY (Node.js) or os.environ["CHAINAWARE_API_KEY"] (Python). Never commit the key to version control.

Tool Reference¶

predictive_fraud - fraud probability + AML forensics
predictive_fraud_batch - batch fraud detection; returns job_id immediately
predictive_behaviour - behavioural profiling + intent prediction
predictive_behaviour_batch - batch behavioural profiling; returns job_id immediately
predictive_rug_pull - smart contract rug pull risk
token_rank_list - ranked token list by holder community strength
token_rank_single - single token deep-dive + top holders
credit_score - crypto trust score (1-9) for lending and credit decisions
check_job_status - check batch job progress
get_job_results - retrieve completed batch job results
run_token_audit - deep smart contract audit (get-or-create)
get_token_audit_result - poll for completed audit results
agents_trust_score_list - list ERC-8004 agents with trust scores
agents_trust_score_single - full trust profile for a single agent