Skip to main content

​
Overview

The AgentScore MCP (Model Context Protocol) server exposes trust lookups as tools that AI agents and assistants can call directly. This lets agents check wallet reputation before transacting, without custom API integration code.

​
Installation

npm install -g @agentscore/mcp

​
Configuration

Add AgentScore to your MCP client configuration (e.g., Claude Desktop, Cursor, or any MCP-compatible host): 
{
  "mcpServers": {
    "agentscore": {
      "command": "agentscore-mcp",
      "env": {
        "AGENTSCORE_API_KEY": "sk_live_abc123..."
      }
    }
  }
}

​
Available tools

​
agentscore_reputation

Look up a wallet’s trust score and reputation. Parameters:
  • address (string, required) — EVM wallet address
  • min_grade (string, optional) — Minimum grade for the decision
  • min_transactions (number, optional) — Minimum transaction count
Example usage by an agent:
“Check the reputation of 0xdb5aa553feeb2c3e3d03e8360b36fb0f7e480671 with a minimum grade of B”
Returns: 
{
  "address": "0xdb5aa553feeb2c3e3d03e8360b36fb0f7e480671",
  "score": 82.4,
  "grade": "B",
  "decision": {
    "allow": true,
    "reasons": ["sufficient_transaction_history", "recent_activity"]
  }
}

​
agentscore_batch

Score multiple wallets at once. Parameters:
  • addresses (string[], required) — Array of EVM wallet addresses (max 100)
  • min_grade (string, optional) — Minimum grade threshold
  • min_transactions (number, optional) — Minimum transaction count

​
agentscore_health

Check API status and data freshness. Returns: 
{
  "status": "healthy",
  "model_version": "v1",
  "last_sync_time": "2026-03-08T17:15:00Z"
}

​
Use cases

​
Agent-to-agent trust

An AI agent can verify the reputation of another agent’s wallet before initiating a transaction: 
Agent: Before paying 0xabc... for the weather data API,
       let me check their AgentScore reputation.

[calls agentscore_reputation with address="0xabc...", min_grade="C"]

Agent: The wallet has a score of 79 (grade C) and is allowed.
       Proceeding with payment.

​
Gateway pre-screening

An agent acting as a gateway can screen incoming requests: 
Agent: Incoming request from wallet 0xdef...
       Checking trust score before routing.

[calls agentscore_reputation with address="0xdef...", min_grade="B"]

Agent: Score is 31 (grade D), below the B threshold.
       Rejecting the request.

​
Batch pre-filtering

An agent managing a marketplace can filter a list of applicants: 
Agent: I have 20 wallets requesting access to the marketplace.
       Let me batch-check them all.

[calls agentscore_batch with addresses=[...], min_grade="C", min_transactions=3]

Agent: 14 wallets passed, 6 were denied. Proceeding with approved wallets.

​
Running locally

For development and testing: 
AGENTSCORE_API_KEY=sk_test_... agentscore-mcp --stdio
The server communicates over stdio using the MCP protocol. Connect any MCP-compatible client to test tool calls interactively.