Loading ToolBench
Read-only on-chain intelligence for AI agents on Base blockchain. Provides access to address balances, token metadata, gas prices, and transaction status without requiring API keys.
Base Intel MCP provides 5 read-only blockchain tools with complete input schemas and reasonable descriptions. All tools have proper JSON Schema definitions with required fields and type constraints. However, parameter descriptions are sparse (only 3 of 15 parameters have descriptions in the schema itself; others are injected via DK_AD lookup), and output schemas are entirely undocumented. Error handling is present but generic. Tool naming is clear and verb-based (address_report, token_info, token_balance, gas_now, tx_status), following the get_/list_ convention reasonably well. The server implements input validation (isAddr, isHash regex checks) and returns structured objects, but lacks guidance on error recovery and does not document return field types for LLM planning.
One-call snapshot of a Base address: ETH balance, USDC balance, tx count, and whether it is a contract.
Current Base network gas price (gwei and wei).
ERC-20 token balance of a holder on Base (formatted + raw).
ERC-20 token metadata on Base: name, symbol, decimals, total supply.
Look up a Base transaction by hash: status, from, to, value, block, gas used.
Output schemas are completely undocumented. LLMs cannot plan downstream tool calls or extract fields without seeing return types. Functions return structured objects (e.g., {address, network, eth_balance, usdc_balance, tx_count, is_contract}) but no schema is published.
Parameter descriptions are injected via a lookup table (DK_AD) rather than embedded in the schema. Only 3 of 15 parameters have descriptions visible in TOOL_DEFS; the rest rely on runtime injection. This is fragile and makes the schema incomplete for static analysis.
Error messages are generic ToolError exceptions (e.g., 'Invalid address. Expected 0x + 40 hex chars.') but do not guide recovery. When RPC fails, the error 'Base RPC unavailable: ...' does not tell the LLM whether to retry, wait, or try a different tool.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | 2026-07-28+ | v2 |
gas_now has no parameters but its description is vague ('Current Base network gas price (gwei and wei).'). It does not explain when to call it or what the output structure is.
No pagination or result limits documented. While these tools return single objects or small results, the server does not declare limits or explain behavior if a token has millions of holders.