Loading ToolBench
A voice-first automation agent for the Cardano blockchain. Provides MCP tools for wallet management, transaction sending, staking operations, and blockchain interaction through an AI assistant interface.
This server exhibits severe quality gaps across nearly all dimensions. Nine tools are present but definitions are incomplete, descriptions are vague or missing critical context, and schemas lack proper structure. The Cardano wallet tools (openCardanoWallet, openMetamaskWallet, SendTransactions, StakingTransaction) use UI automation with pyautogui and lack input validation, error handling, and security considerations. The Masumi tools (list_agents through get_job_full_result) have better descriptions but incomplete parameter annotations. No tool documents output schemas. Error handling is minimal, most tools return raw error strings rather than actionable recovery guidance. The server mixes blockchain operations with agent delegation without clear separation of concerns.
Sends a transaction on the Cardano blockchain by automating the wallet UI. Fills in the recipient address, amount, confirms the transaction, waits for user password entry, and submits.
Performs a staking operation on the Cardano blockchain by automating the wallet UI to select a stake pool, confirm delegation, and submit the transaction with user password verification.
Checks the status of a job on a Masumi agent. Returns the current job status and progress information.
Retrieves the required input schema for a specific Masumi agent using its URL. Returns a JSON string detailing the input schema or an error message.
Retrieves the complete result of a finished job from a Masumi agent. Returns the full output and results of the completed job.
Hires a Masumi agent using its URL: starts a job and initiates payment. Use 'get_agent_input_schema' first to understand the required structure for 'input_data'. The input_data parameter should contain values explicitly provided by the user, not automatically generated by the assistant.
Wallet automation tools (openCardanoWallet, openMetamaskWallet, SendTransactions, StakingTransaction) use UI automation via pyautogui without input validation, error handling, or confirmation patterns. SendTransactions and StakingTransaction have IRREVERSIBLE risk but lack dry-run or confirmation_request pattern support.
No tool documents output schema. LLMs cannot determine what fields to expect, making composition and chaining impossible. Required per the rubric (100% of A+ tools have documented return types).
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 43 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 33 | - | v1 |
Lists available agents on the configured Masumi network from the registry. Returns a JSON string of agent entries, including 'agentIdentifier' and 'apiBaseUrl'.
Opens the Cardano wallet extension in Chrome browser by pressing Win, typing 'chrome', and navigating to the Cardano wallet extension URL.
Opens the Metamask wallet extension in Chrome browser by pressing Win, typing 'chrome', and navigating to the Metamask wallet extension URL.
Descriptions for wallet tools are vague and lack WHEN guidance. 'Opens the Cardano wallet extension in Chrome browser...' does not explain when the LLM should call it or what prerequisites exist. Descriptions are 35 chars (below 100-char minimum for clarity).
SendTransactions and StakingTransaction accept string parameters (Address, Amount, pool_name) without type constraints, enums, or format validation. LLMs can pass invalid addresses or amounts; no guidance on format.
Error handling across all tools is minimal. Most return raw error strings ('Error: ...') without categorization (retryable vs fatal), actionable next steps, or recovery guidance. Does not follow error-classification or recovery-guide patterns.
Masumi tools (list_agents, get_agent_input_schema, hire_agent, check_job_status, get_job_full_result) accept parameters like agent_identifier and api_base_url without documenting format, constraints, or examples. Parameter descriptions are sparse (30-50 chars vs 72-char baseline).
No evidence of tool composition or chaining support. Wallet tools do not output identifiers needed by subsequent operations. Masumi tools return JSON strings rather than structured objects, forcing LLM parsing of free-text output.
UI automation tools (openCardanoWallet, openMetamaskWallet) have platform-specific dependencies (pyautogui, pytesseract) and hardcoded Win key interaction. No graceful degradation, timeout handling, or cross-platform support documented.
hire_agent tool accepts arbitrary 'input_data' as an object parameter without schema validation. Description warns 'should contain values explicitly provided by the user', but no client-side or server-side enforcement exists. Vulnerable to prompt injection.
No permission gates or scope declarations. Wallet tools can send transactions without explicit authorization checks. No audit trail documenting who initiated what transaction at what time.