Purpose
AgentRadar MCP is the MCP server for AgentRadar, providing on-chain trust oracle capabilities for the agent economy. For ERC-8004 and x402 agents on the Base network, it provides trust scores, scam detection, and EAS on-chain attestations, helping users directly verify, score, and attest to the trustworthiness of AI agents in Claude, Cursor, or any MCP-compatible client.
Main Capabilities
This MCP server provides 18 tools for verifying, scoring, and attesting to the trustworthiness of AI agents. According to the materials, it can support the following typical operations:
- Verify an agent address (for example: "Verify this agent: 0x742d35Cc6634C0532925a3b844Bc9e7595f2bD28")
- Check whether a wallet is a known scam wallet (for example: "Is 0xdead...beef a known scam wallet?")
- Compare two agents and determine which is more trustworthy
- Explain why an agent received a score of 45/100
- Write an on-chain attestation for a verified agent
The tools explicitly listed in the materials include:
- verify_agent: $0.005 per call (x402) or free with an operator key
- attest_agent: $5.00 per attestation (x402) or free with an operator key
- get_score: free
- Admin endpoints: require an operator key
Usage
This package is a lightweight client wrapper around the public AgentRadar API. It can be run via npx @agentradar/mcp, the code is auditable in this repository, and it uses the MIT license.
Integrating with Cursor: add it to the project's .cursor/mcp.json. Integrating with Claude: on macOS, add it to ~/Library/Application Support/Claude/claude_desktop_config.json; on Windows, add it to %APPDATA%\Claude\claude_desktop_config.json.
The materials also mention Local development and Environment Variables sections, but do not provide the specific configuration contents.
Applicable Scenarios
Suitable for scenarios that require verifying the other party's trustworthiness during AI agent interactions, such as checking in Claude or Cursor whether an agent address is trustworthy, whether it is a scam wallet, comparing the trustworthiness of multiple agents, understanding the basis for a score, and generating on-chain attestations for verified agents.
Limitations
The materials do not provide the complete installation command, the specific contents of environment variables, local development steps, details of supported on-chain networks (other than Base), details of the scoring algorithm, API rate limits, or availability guarantees. Some tools (verify_agent, attest_agent) are charged per use, and Admin endpoints require an operator key.
