> ## Documentation Index
> Fetch the complete documentation index at: https://docs.credprotocol.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Agent Skill
> Instructions that AI agents receive when connecting to the Cred Protocol MCP server
# Agent Skill Instructions
When an AI agent connects to the Cred Protocol MCP server, it automatically receives the following instructions during initialization. These guide the agent on tool selection, score interpretation, and best practices.
These instructions are delivered via the MCP `instructions` field — no extra configuration needed. You can also use this page as a reference when building custom agents.
***
## When to Use These Tools
Use Cred Protocol tools when users ask about:
* Wallet creditworthiness, trust, or risk assessment
* Financial health or portfolio analysis of an Ethereum address
* Identity verification or attestation checks
* Sybil detection (is this a real person or a bot?)
* DeFi lending eligibility or underwriting decisions
* Comparing multiple wallets or batch assessments
* Transaction flow and counterparty analysis
* AI agent discovery, reputation, or on-chain registration
## Tool Selection Guide
**Start here — pick the right tool for the job:**
| User Intent | Tool | Notes |
| -------------------------------------------- | --------------------------- | -------------------------------------- |
| "Is this wallet trustworthy?" | `get_credit_score` | Add `include_factors: true` for detail |
| "Compare these wallets" | `get_credit_scores_batch` | Individual scores, no aggregation |
| "Give me one score for this multi-sig" | `get_aggregate_score` | Single combined score |
| "Tell me everything about this wallet" | `get_financial_summary` | Full report with assets, DeFi, events |
| "Full multi-chain credit report" | `get_comprehensive_report` | Per-chain breakdowns + percentiles |
| "Quick summary, no chain details" | `get_summary_report` | Lighter than comprehensive |
| "What's happening on Ethereum specifically?" | `get_chain_report` | Single-chain deep dive |
| "Quick chain stats" | `get_chain_summary` | Single-chain aggregated metrics |
| "How much is this wallet worth?" | `get_portfolio_value` | Total USD across all chains |
| "What does this wallet hold on Base?" | `get_chain_portfolio_value` | Chain-specific USD value |
| "What's the asset breakdown?" | `get_portfolio_composition` | Tokens, DeFi, stablecoins, debt |
| "What identity does this wallet have?" | `get_identity_attestations` | ENS, Gitcoin Passport, POAPs, etc. |
| "Is this a bot or a real person?" | `get_sybil_score` | Sybil detection with indicators |
| "Show me this wallet's transaction network" | `get_transaction_graph` | Token transfer graph (nodes + edges) |
| "How many agents are registered?" | `get_agent_count` | Total in ERC-8004 registry |
| "List all agents" | `list_agents` | Paginated agent listing |
| "Find agents that do X" | `search_agents` | Search by name or description |
| "Look up agent #42" | `get_agent` | Single agent info by ID |
| "Rate this agent on-chain" | `submit_agent_reputation` | Submits credit/sybil/identity scores |
| "What's this agent's reputation?" | `get_agent_reputation` | Aggregated on-chain feedback |
| "Did my reputation submission go through?" | `get_reputation_status` | Track async submission |
## Escalation Patterns
**Wallet assessment (most common):**
Call `get_credit_score` with `include_factors: true` — this answers 80% of questions.
If the user wants the full picture, call `get_financial_summary` or `get_comprehensive_report`.
Only call `get_identity_attestations`, `get_portfolio_value`, or `get_sybil_score` if the user needs those specific details — the financial summary already includes identity and asset data.
**Deep dive on a wallet:**
1. `get_credit_score` for the headline number
2. `get_comprehensive_report` for the full multi-chain breakdown
3. `get_sybil_score` if trust/humanity is in question
4. `get_transaction_graph` to visualize the wallet's network
**Agent reputation workflow:**
1. `search_agents` or `list_agents` to find the agent
2. `get_agent` to confirm details
3. `get_agent_reputation` to read existing on-chain feedback
4. `submit_agent_reputation` to add new feedback (costs 10 CU, async)
5. `get_reputation_status` to track the on-chain submission
## Authentication
All tools require a `CRED_API_KEY` to access the live Cred Protocol API. All responses include `source: "live"` indicating real on-chain data.
Generate your API key from the [Cred Protocol Dashboard](https://app.credprotocol.com/dashboard).
## Understanding Credit Scores
Scores range from 300 to 1000 (like traditional credit scores):
| Range | Label | Meaning |
| -------- | --------- | ------------------------------------------- |
| 920-1000 | Excellent | Top-tier on-chain reputation. Minimal risk. |
| 840-919 | Very Good | Strong history, diversified activity. |
| 750-839 | Good | Solid track record, moderate experience. |
| 640-749 | Fair | Limited history or some risk signals. |
| 300-639 | Low | New wallet, thin file, or negative events. |
**Score factors** (available with `include_factors: true`):
* **Borrowing History** — Loan repayment track record across DeFi protocols
* **Wallet Composition** — Asset diversity and stablecoin ratio
* **Wallet Health** — Collateralization ratios and liquidation distance
* **Interactions** — Protocol breadth and ecosystem participation
* **Trust** — Identity attestations (ENS, Gitcoin Passport, POAPs)
* **New Credit** — Recent changes in borrowing behavior
## Understanding Sybil Scores
The `get_sybil_score` tool returns a score from 0-100 with a risk level:
| Risk Level | Score | Meaning |
| ---------- | ------ | ------------------------------------------- |
| low | 0-25 | Likely a unique human with organic activity |
| medium | 26-50 | Some suspicious indicators, needs context |
| high | 51-75 | Multiple sybil indicators present |
| critical | 76-100 | Very likely a sybil/bot account |
**Key indicators to highlight:**
* `wallet_age_days` — Older wallets are more trustworthy
* `transaction_time_entropy` — Higher entropy = more human-like timing
* `identity_attestations` — More attestations = harder to fake
* `count_unique_counterparties` — Diverse interactions suggest real usage
## Reading Financial Summaries
Key fields in the financial summary report:
| Field | Meaning |
| ------------------------------------ | ----------------------------------------------- |
| `net_worth_usd` | Total assets minus total debt |
| `total_asset_usd` / `total_debt_usd` | Gross positions |
| `total_collateral_usd` | Assets locked as collateral in DeFi |
| `count_transactions` | Total on-chain transactions (activity proxy) |
| `count_active_loans` | Current open DeFi loans |
| `count_liquidations` | Historical liquidation events (red flag if > 0) |
| `global_percentiles` | How this wallet compares to the population |
## Identity Attestations
Attestations are verified on-chain identity credentials:
| Attestation | What It Proves |
| ---------------- | -------------------------------------------- |
| ENS Name | Owns an Ethereum Name Service domain |
| Basename | Owns a Base network name (username.base.eth) |
| Gitcoin Passport | Passed humanity verification (score >= 20) |
| POAPs | Attended real-world or virtual events |
| Worldcoin | Verified unique human via biometric proof |
| BrightID | Verified through social graph analysis |
More attestations generally indicate a more established, trustworthy identity.
## Transaction Graphs
The `get_transaction_graph` tool returns a tripartite graph:
* **Nodes** with `group: "wallet"` are Ethereum addresses
* **Nodes** with `group: "token"` are ERC-20 tokens (USDC, WETH, etc.)
* **Edges** with `label: "SENT"` go from wallet to token
* **Edges** with `label: "RECEIVED"` go from token to wallet
This is useful for visualizing money flow, identifying major counterparties, and spotting wash trading patterns.
## Agentic Reputation (ERC-8004)
The agent tools interact with the ERC-8004 Identity and Reputation Registries on Base (chain ID 8453).
**Identity Registry** — where agents are registered as ERC-721 tokens:
* Each agent has an `agent_id` (token ID), `owner` address, and `metadata` (name, description, image)
* Use `get_agent_count`, `list_agents`, `search_agents`, `get_agent` to browse
**Reputation Registry** — where feedback about agents is stored on-chain:
* `submit_agent_reputation` computes credit score, sybil risk, and identity count for a given address, then writes the results on-chain as feedback for the specified agent
* `get_agent_reputation` reads the aggregated feedback summary
* Submissions are async — use `get_reputation_status` to track
You can combine agent tools with wallet tools: look up an agent's owner via `get_agent`, then score the owner with `get_credit_score` to assess the human behind the agent.
## Supported Chains
Data is aggregated across 10 EVM networks:
| Chain | ID | Notes |
| --------- | ------ | ------------------------------ |
| Ethereum | 1 | Primary chain, richest data |
| Optimism | 10 | L2 |
| BSC | 56 | Binance Smart Chain |
| Polygon | 137 | L2 |
| Base | 8453 | Coinbase L2, ERC-8004 registry |
| Arbitrum | 42161 | L2 |
| Celo | 42220 | Mobile-first chain |
| Avalanche | 43114 | — |
| Scroll | 534352 | zkEVM L2 |
| Linea | 59144 | Consensys zkEVM L2 |
## Input Format
All address-based tools accept either:
* **Ethereum addresses**: `0x742d35Cc6634C0532925a3b844Bc9e7595f32345` (42 characters, hex)
* **ENS names**: `vitalik.eth`, `username.base.eth`
Agent tools use integer `agent_id` values (starting at 1).
## Best Practices
A single `get_credit_score` with `include_factors: true` answers most questions. Only escalate to `get_comprehensive_report` if the user needs the full breakdown.
Use `get_credit_scores_batch` instead of looping `get_credit_score` for multiple addresses.
Translate scores and data into actionable insights. "This wallet has an Excellent score of 945 with no liquidation history — it's a strong candidate for undercollateralized lending."
Highlight liquidations, high debt-to-asset ratios, low scores, missing identity attestations, or high sybil scores.
A 750 is "Good" — don't call it average. A 640 is "Fair" — don't call it bad. Use the labels.
Credit score + sybil score + identity attestations gives a comprehensive trust picture. Agent lookup + owner credit score assesses the human behind an agent.
After `submit_agent_reputation`, always inform the user the submission is async and provide the `submission_id` for tracking with `get_reputation_status`.