Loading ToolBench
MCP server for ClawMind, an AI-powered Web3 risk analysis platform that integrates with 0G Storage, 0G Compute, and on-chain analysis registry. Provides tools for analyzing Web3/AI projects, managing memory indices, and retrieving analysis reports.
ClawMind exposes 9 tools via HTTP with basic descriptions and input schemas. However, critical gaps undermine quality: (1) Most tool descriptions lack actionable context for LLM selection, they describe WHAT but not WHEN or WHY to call them. (2) Input schemas present but sparse: many parameters lack type constraints (e.g., 'task' is just string with no enum/pattern); optional parameters not clearly marked. (3) Output schemas completely undocumented, LLMs cannot plan downstream calls or extract chaining IDs. (4) No error handling guidance, tools return HTTP status codes with no recovery hints. (5) Tool composition issues: analyze_web3_project is a mega-tool combining analysis, memory retrieval, and on-chain registration in one call, violating single-responsibility. (6) Security: 'authorization' parameter in get_debug_info suggests token passing in params (anti-pattern). Average tool description ~150 chars (within baseline 194), but quality is generic. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Median tool score ~45, consistent with community baseline.
Analyzes a Web3/AI project by running a multi-step agentic pipeline. Accepts a task description, runs semantic memory retrieval, executes specialized analysis agents, and returns a structured report with score, recommendation, risks, opportunities, and architecture insights.
Generates a formatted PDF report of an analysis result. Includes task description, score, recommendation, risk map, opportunities, architecture, next steps, evidence log, and infrastructure receipts.
Polls the status of an ongoing analysis task by taskId. Returns current step, progress, and final result when complete. Also returns infrastructure status (compute, storage, on-chain registry configuration) when no taskId is provided.
Returns configuration status and diagnostic information for debugging infrastructure issues. Shows network, compute, storage, and on-chain registry configuration without revealing secrets. Includes auto-diagnosis of common configuration problems.
Returns comprehensive evidence for hackathon judges on a single read-only page. Includes project info, network configuration, 0G integration status, latest on-chain analysis, analysis metrics, memory statistics, and MCP usage tracking.
Retrieves all memory records from the persistent memory store. Returns an array of memory records with summaries, risks, embeddings, and metadata.
Output schemas completely undocumented. LLMs cannot infer what fields to expect from any tool response, blocking downstream tool chaining and forcing exploratory calls.
Descriptions lack WHEN/WHY context. 'Analyzes a Web3/AI project...' tells WHAT but not when to call it vs. get_analysis_status or retrieve_report_from_storage. LLMs cannot disambiguate.
Input schemas lack type constraints and enums. 'source' parameter has enum ['mcp','web'] but 'task' is bare string with no pattern/minLength validation. 'format' in get_openclaw_manifest has enum but others lack constraints.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 45 | 2026-07-28+ | v2 |
Retrieves the OpenClaw pipeline manifest in YAML or JSON format. JSON format includes live 0G evidence (network, compute, storage, on-chain registry status, semantic memory configuration) and manifest validation results.
Retrieves a previously saved analysis report from 0G Storage using a storage URI or root hash. Returns the full report data including summary, score, recommendation, risks, opportunities, and evidence.
Verifies the integrity of the latest on-chain analysis by reading from the Analysis Registry contract. Returns the on-chain data, contract details, and integrity checks (hash format, score range, valid recommendation, storage URI presence, submitter format, operator signature verification).
No error handling guidance. Tools return HTTP status codes with no recovery hints. E.g., if analyze_web3_project fails, LLM has no guidance on retry vs. user-fixable vs. fatal.
Security anti-pattern: 'authorization' parameter in get_debug_info suggests token passing in tool parameters. Credentials must use server-side injection via environment variables, not params.