Both tools have minimal descriptions (under 100 chars) and lack parameter descriptions entirely. Input schemas are present but bare, only type and description for a single string parameter, no constraints or validation guidance. No output schema documented. Tools expose dangerous capabilities (arbitrary Python execution in Blender) with no error handling guidance, confirmation patterns, or permission gates. Naming is verb-based but descriptions fail to explain WHEN to use each tool or what distinguishes eval from exec. No idempotency, no pagination, no structured output guidance.

• Add parameter descriptions: 'expression' should explain it must be a valid Python expression (single line, returns a value). 'code' should explain it can be multi-line statements (loops, conditionals, assignments).
• Document output schema: specify whether output is plaintext, JSON, or raw Python repr. Clarify that exceptions return traceback strings, not structured error objects.
• Expand tool descriptions to 100-200 chars each. Explain: eval_python returns the result of evaluating an expression; exec_python executes statements and returns None or side effects. When to use each.
• Add a dry_run parameter (boolean, default false) to exec_python. When true, parse and validate code without executing. Prevents accidental destructive operations.
• Implement structured error responses: return {"success": false, "error": "SyntaxError: invalid syntax", "line": 5, "suggestion": "Check indentation"} instead of raw tracebacks. Enables LLM recovery.