Server has 9 tools with basic descriptions and input schemas, but significant quality gaps. All tools lack parameter descriptions, critical for LLM reasoning. Output schemas are undocumented (Record<string, unknown>). Two write tools (arcis_deposit, arcis_withdraw) expose private_key as a parameter, violating secret-injection pattern. Descriptions are terse (10-50 chars) and lack context on when/why to use each tool. No error handling guidance, no pagination support, no idempotent hints. Tool naming is consistent (verb_noun) but descriptions don't explain dependencies or prerequisites.

• Move private_key handling to server-side secret injection. Accept an optional 'signer_address' parameter instead, and resolve the private key from environment or secure vault inside the tool implementation.
• Add 20-50 character descriptions to every parameter. Example: 'agent_address: Ethereum address (0x...) to check vault balance for. Must be a valid checksummed address.'
• Document output schemas explicitly. For arcis_vault_status, specify: {tvl: string (USD), exchange_rate: string, supply: string (shares), capacity: string (USD), reserve: string (USD), deployed: string (USD)}.
• Add a dry-run parameter to arcis_deposit and arcis_withdraw. When dry_run=true, return estimated gas, slippage, and final amounts without executing. This lets agents preview before committing.
• Expand tool descriptions to 100-200 chars. Include: what data is returned, when to call it (e.g., 'Call before deposit to check available capacity'), and any prerequisites (e.g., 'Requires agent_address from arcis_vault_balance').