Loading ToolBench
MCP server for AgentOracle — claim verification for AI agents. Verify before you act. x402-native on Base, SKALE, Stellar.
Three tools with clear, descriptive names and good descriptions (150-200 chars each). All tools have input schemas with typed parameters and descriptions. However, output schemas are not documented in the source code, responses are returned as raw JSON without explicit schema definitions. Error handling is minimal (basic HTTP status checks). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk profiles. Parameter constraints are present (maxLength) but could be more explicit about format and validation rules. Tools follow verb_noun naming (preview, research, deep-research) and descriptions explain WHAT, WHEN, and cost implications, good for LLM selection. Missing: output schema documentation, error recovery guidance, and structured error messages.
Comprehensive deep research using Sonar Pro. Returns expert-level analysis with 10-15 detailed facts, in-depth synthesis, cited sources, and confidence score. Costs $0.10 USDC via x402. Use for complex topics requiring thorough analysis.
Free research preview — get a quick summary, 2 key facts, and confidence score for any query. No payment required. Use this to test queries before committing to a paid research call. Rate limited to 10/hour.
Real-time research on any topic. Returns structured JSON with summary, key facts, cited sources, and confidence score. Costs $0.02 USDC via x402 (Base or SKALE gasless). If a wallet is configured, payment is handled automatically.
Output schemas not documented. Tools return raw JSON without explicit schema definitions. LLMs cannot plan downstream tool calls or extract fields reliably.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). All three tools are READ_ONLY but this is not declared in the tool definition, only in metadata comments.
Error handling returns raw HTTP status and JSON without recovery guidance. 'Preview failed (HTTP 402)' tells the LLM nothing about what to do next (retry, ask user for wallet, etc.).
Parameter descriptions lack explicit format/constraint details. 'Natural language research question' is clear, but no guidance on what makes a valid query, expected response time, or rate limit behavior.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 65 | 2026-07-28+ | v2 |
No pagination or result-limiting guidance in descriptions. Tools return 'structured JSON' but no documentation of field names, array sizes, or how to handle large result sets.