Loading ToolBench
Credit bureau for AI agent wallets — trust scores from public blockchain data. Exposes wallet trust data to AI agents via Model Context Protocol. Calls the public AgentKarma REST API.
AgentKarma MCP server demonstrates solid tool design with clear naming conventions (verb_noun pattern), comprehensive descriptions (avg 180 chars), and well-structured input schemas using Zod validation with regex patterns for EVM addresses. All 7 tools have descriptions exceeding 100 chars and parameter types are properly defined. However, output schemas are not formally documented, responses are returned as JSON text without explicit schema declarations. Error handling is basic (404 detection only). Tool composition is strong: batch_trust_scores and list_wallets support pagination; get_wallet_trust_signals combines multiple data sources. Missing: tool annotations (readOnlyHint/destructiveHint), detailed error recovery guidance, and output schema documentation.
Get AgentKarma platform statistics: total wallets indexed, transactions scored, feedback count, score distribution by tier, and database usage.
Look up trust scores for multiple wallet addresses at once (max 100). Returns scores with tier and breakdown for found wallets, and a list of addresses not found. Use this when comparing multiple wallets or checking a batch of counterparties.
Get the trust score (0-100) for an AI agent wallet address. Returns a weighted score based on 7 signals: loyalty (30%), activity (18%), diversity (16%), feedback (15%), volume (10%), recency (6%), age (5%). Includes tier (HIGH/MEDIUM/LOW/MINIMAL), percentile rank, and signal-by-signal breakdown. Use this as your primary tool for trust decisions before transacting.
Deep trust analysis for a wallet. Returns the trust score with breakdown PLUS recent transaction history showing counterparties, amounts, and roles (payer/recipient). Use this when you need to understand WHY a wallet has its score — e.g., before a high-value transaction where you want to see the wallet's actual on-chain behavior.
Browse indexed AI agent wallets. Filter by data source (erc8004 = registered agents, x402 = payment activity). Returns a paginated list sorted by recent activity.
Output schemas not formally documented. Tools return JSON text via jsonText() helper without declaring expected response structure. LLMs cannot plan downstream operations or validate field presence.
Tool annotations missing. No readOnlyHint, destructiveHint, or idempotentHint declared. submit_feedback is a WRITE operation but not marked as such in tool metadata.
Error handling is minimal. Only 404 detection; no guidance on retryability, user-fixable errors, or recovery steps. Error messages like 'API 500: ...' provide no actionable next steps for LLM.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 64 | 2026-07-28+ | v2 |
Get full wallet identity and metadata: source (erc8004/x402), chain, ERC-8004 agent ID, trust score with tier, transaction and feedback counts, and activity timestamps. Use this when you need wallet details beyond the trust score — e.g., "is this an ERC-8004 registered agent?" or "how many transactions has this wallet made?"
Submit feedback for a wallet after a transaction. Rate the experience 1-5 stars. This contributes to the wallet's trust score over time. Use this after completing a transaction to help build the trust network.
agentkarma_stats has empty input schema (no parameters). Description is generic (75 chars). Unclear what 'database usage' means or how to interpret stats.
list_wallets 'source' parameter enum values (erc8004, x402, both) lack explanation in description. LLM cannot infer what these protocols represent without external knowledge.