How to Use AgentScore in MCP Tools

If your assistant can already call MCP tools, you can make wallet screening part of the tool loop. That means an agent can check a wallet before it pays it, routes work to it, or treats it as trustworthy.

The published package for this is djd-agent-score-mcp. It exposes wallet-scoring tools over MCP so a host like Claude Desktop or Cursor can call them directly without you building a custom integration first.

The Fastest Setup

Add the server to your MCP client config:

{
  "mcpServers": {
    "djd-agent-score": {
      "command": "npx",
      "args": ["-y", "djd-agent-score-mcp"],
      "env": {
        "DJD_API_KEY": "djd_live_..."
      }
    }
  }
}

If you leave out DJD_API_KEY, the free tools still work with the free-tier limits. Adding a key unlocks the paid tools and higher usage.

What The MCP Server Exposes

get_score for a basic score, confidence, tier, and recommendation
get_full_score for dimension breakdowns, sybil flags, and improvement paths
get_score_history for trend analysis
batch_score for scoring several wallets together
Free read tools like leaderboard and economy metrics

That is enough to make trust checks part of the same agent loop that is already reading tools, calling APIs, or deciding whether to spend money.

How Agents Actually Use It

The useful pattern is not “ask the model to invent trust.” The useful pattern is “give the model a concrete tool that returns a score, confidence, and recommendation.”

User: "Should I pay this wallet for a tool call?"

Agent:
1. Calls get_score(wallet)
2. Reads score, confidence, tier, recommendation
3. Decides allow / review / avoid
4. Explains the decision in plain English

That keeps the trust signal grounded in a tool response instead of turning it into a vibes-based model guess.

What This Is Good For

Before x402 payments: check the wallet before paying a new service
Before payouts: inspect an external destination wallet before sending funds
Before delegation: ask whether a wallet looks strong enough to receive work or access
During investigations: use history or full score tools to explain a suspicious wallet

The Main Point

MCP is a distribution surface for trust checks. It lets you put the wallet-screening step inside the same place the agent is already reasoning and acting.

Use the published MCP server or start with the API docs.

npx -y djd-agent-score-mcp

Open Docs

The MCP server gives AgentScore a natural distribution path into Claude, Cursor, and other MCP-capable clients. It is one of the easiest ways to make trust checks part of an agent tool loop.