Loading ToolBench
AI Agent interface for the Metaplex Genesis Protocol on Solana Devnet. Helps users create tokens, fetch accounts, and interact with bonding curves through an MCP server.
Single tool 'execute_mcp_tool' has a reasonable name and description, but the schema is problematic. The 'arguments' parameter accepts a JSON string rather than structured objects, forcing the LLM to serialize/deserialize and losing type safety. No output schema is documented. The tool acts as a generic wrapper around arbitrary Metaplex tools rather than exposing those tools directly to the agent, which violates the composition pattern, agents cannot reason about what each underlying tool does. No error handling guidance, no parameter validation rules, and no documentation of what the underlying tools return.
Execute a Metaplex Genesis MCP tool
Tool wraps arbitrary MCP tools instead of exposing them directly. LLM cannot reason about individual tool semantics, capabilities, or error modes. Violates single-responsibility and composition patterns.
'arguments' parameter accepts free-form JSON string instead of structured schema. Loses type safety, forces LLM to manually serialize, and prevents schema validation. Should use oneOf or discriminated union to expose actual tool schemas.
No output schema documented. LLM cannot predict what fields to expect or plan downstream operations. Tool description does not explain return structure.
No error handling guidance. Tool catches errors and returns generic strings like 'Tool execution failed with an unknown error', does not tell LLM what to do next or whether to retry.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | 2026-07-28+ | v2 |
Parameter descriptions lack validation rules. 'tool_name' and 'arguments' have no constraints on format, length, or allowed values. LLM has no guidance on what tool names are valid.