Give Your AI Agent a Reputation Analyst: Cred Protocol's Full MCP Toolkit

Why Your AI Agent Needs Reputation Intelligence

AI agents are starting to make real decisions — gating access, delegating tasks to other agents, routing payments, evaluating counterparties. But most agents today are flying blind when it comes to the reputation of the wallets and agents they interact with.

Can this wallet be trusted? Is it a bot? Does it have any verified identity? What's its on-chain track record? Has this agent been rated positively by other agents? These are questions that matter for any agent operating in DeFi, payments, or multi-agent systems — and until now, answering them required stitching together multiple APIs, parsing raw blockchain data, and building your own scoring logic.

Cred Protocol's MCP services change that. We've packaged our full reputation intelligence stack — the same infrastructure used by protocols like 3Jane and integrated into Coinbase's x402 — into 21 tools that any AI agent can discover and use through the Model Context Protocol.

What's in the Toolkit

The MCP server exposes five categories of tools, all returning live on-chain data aggregated across 10 EVM networks.

Agent Reputation (ERC-8004)

The headline capability. Seven tools provide full access to the ERC-8004 Identity and Reputation Registries on Base — enabling AI agents to discover, evaluate, and rate other AI agents on-chain.

search_agents and list_agents let an agent browse the registry. get_agent returns details for a specific agent. submit_agent_reputation computes a reputation assessment — credit score, sybil risk, and identity count — for a given address and writes the results on-chain as feedback. get_agent_reputation reads the aggregated feedback. It's a full reputation loop — agents assessing agents, with the results stored immutably on Base.

This is the foundation of trust in multi-agent systems. Before your agent delegates a task, it can check who it's dealing with.

Identity and Sybil Detection

Two tools that answer the most fundamental trust questions. get_identity_attestations checks for verified credentials: ENS names, Basenames, Gitcoin Passport, POAPs, Worldcoin, and BrightID. More attestations from independent providers means a more established, harder-to-fake identity.

get_sybil_score goes further — analyzing transaction timing entropy, counterparty diversity, funding source patterns, and wallet age to produce a 0-100 sybil risk score. A score of 12 means "almost certainly a real person." A score of 85 means "almost certainly a bot." For any agent that needs to distinguish real users from sybils, this is the signal.

Reputation Scoring

get_credit_score returns a reputation score from 300 to 1000 for any Ethereum address or ENS name — computed entirely from on-chain behavior. Borrowing history, wallet composition, collateralization health, protocol interactions, and identity attestations all feed into the score. It's the quantitative backbone of wallet reputation.

For multi-sig wallets or portfolio-level assessments, get_aggregate_score produces a single combined score across multiple addresses. Need to evaluate a batch of counterparties? get_credit_scores_batch handles that in one call.

On-Chain Activity Reports

When a score isn't enough, the reporting tools go deeper. get_financial_summary returns a complete on-chain profile: net worth, assets, debt, collateral, DeFi positions, transaction history, credit events, and global percentile rankings. get_comprehensive_report adds per-chain breakdowns across all 10 supported networks.

For lighter queries, get_summary_report gives you aggregated metrics without the chain-level detail, and get_chain_report lets you drill into a single network. get_portfolio_composition breaks holdings down by type — tokens, stablecoins, DeFi positions, collateral, and debt — so an agent can understand not just how much a wallet holds, but how it holds it.

Transaction Graph

get_transaction_graph returns a visualization of token transfer relationships — wallets, tokens, and the flow between them. This is useful for identifying major counterparties, spotting wash trading patterns, and understanding the network around any address. Reputation isn't just about a wallet in isolation — it's about who it transacts with.

Getting Started in 2 Minutes

The fastest way to connect is through any MCP-compatible client. No SDK installation. No server to run. Just point your client at the endpoint.

Claude Code

Add to your project's .mcp.json:

{
  "mcpServers": {
    "cred-protocol": {
      "type": "url",
      "url": "https://api.credprotocol.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

That's it. Ask Claude Code "What's the reputation of vitalik.eth?" and it calls the right tools, interprets the results, and responds with a full analysis — score, identity attestations, sybil risk, and recommendations.

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "cred-protocol": {
      "url": "https://api.credprotocol.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Claude Desktop

Same configuration — add it to your MCP settings and Claude can access all 21 tools through natural language.

Direct HTTP

Every MCP tool is also available as a standard REST endpoint at https://api.credprotocol.com/mcp/* for applications that don't use MCP:

# Reputation score
curl -H "Authorization: Bearer YOUR_API_KEY" \
  "https://api.credprotocol.com/mcp/score/vitalik.eth?include_factors=true"

# Sybil detection
curl -H "Authorization: Bearer YOUR_API_KEY" \
  "https://api.credprotocol.com/mcp/identity/vitalik.eth/sybil"

# Search the agent registry
curl -H "Authorization: Bearer YOUR_API_KEY" \
  "https://api.credprotocol.com/mcp/agents/search?q=lending"

# Read an agent's on-chain reputation
curl -H "Authorization: Bearer YOUR_API_KEY" \
  "https://api.credprotocol.com/mcp/agents/1/reputation"

What You Can Build

The toolkit is general-purpose, but some patterns are especially powerful.

Agent-to-Agent Trust Networks

Before your agent delegates a task to another agent, it can look up the target in the ERC-8004 registry, check the owner's reputation score, read on-chain feedback from other agents, and make an informed decision about whether to trust it. After the interaction, it can submit its own reputation feedback — building a decentralized trust graph where agents rate each other based on real experience. This is the foundation of trust in multi-agent systems.

Trust-Gated Access

Gate access to premium features, airdrops, or governance based on wallet reputation. An agent can require a minimum reputation score of 750, at least two identity attestations, and a sybil score below 30 — all checked in a single conversation turn. No custom integration needed. The MCP tools handle the evaluation; the agent handles the logic.

Sybil-Resistant Airdrops and Governance

Before distributing tokens or granting voting power, screen every address for sybil risk. Combine sybil scores, identity attestation counts, and wallet age into a composite eligibility check. Flag wallets with zero attestations, high sybil scores, or freshly created addresses. The agent produces an audit trail of every check — useful for both community trust and regulatory reporting.

Automated Reputation-Based Underwriting

An agent that evaluates DeFi loan applications using reputation signals. Pull the reputation score, check for liquidation history, verify identity attestations, and run sybil detection — all in parallel. The agent can approve, deny, or adjust terms based on the borrower's on-chain reputation, with every data point sourced from live blockchain state.

Counterparty Due Diligence

Screen wallets before onboarding them to your protocol. Combine reputation scoring, sybil detection, identity attestations, and transaction graph analysis into a comprehensive trust assessment. Understand not just a wallet's score, but who it transacts with and how it behaves. The agent produces a complete reputation profile that can inform any access or risk decision.

Built-In Agent Intelligence

When an AI agent connects to the MCP server, it automatically receives skill instructions that guide it on which tool to use for different questions, how to interpret reputation signals, when to escalate from a quick score check to a full on-chain report, and what red flags to highlight. The agent doesn't need training data or custom prompts — the skill instructions are delivered during MCP initialization.

This means an agent connected to Cred Protocol doesn't just have access to reputation data — it knows how to use it. It understands that a score of 750 is "Good," that liquidations are a red flag, that more identity attestations mean higher trust, and that batch scoring is more efficient than individual calls.

The Full Tool Reference

Start Building

Everything you need is live today. Generate an API key from the Cred Protocol Dashboard, add the MCP configuration to your client, and start asking questions about wallets and agents.

For the full setup guide, tool reference, and code examples, see the MCP Services documentation. The agent skill instructions are also available at api.credprotocol.com/mcp/skill.md — useful as a reference when building custom agent prompts.

If your agents are making trust decisions about wallets or other agents, they should be making informed decisions. That's what this toolkit is for.