CutterMCP-plus is a binary analysis tool wrapper around a local Cutter FastAPI plugin. All 20 tools are present with descriptions and input schemas. However, there are consistent gaps in schema completeness, output documentation, and error handling guidance. Naming is action-verb-based and generally clear (list_*, get_*, disasm_*, decompile, rename_*, set_*). Descriptions are mostly adequate (80-150 chars) but lack actionable context for LLM decision-making and error recovery. Parameter schemas are present but lack proper typing constraints (enums, ranges, patterns). Output schemas are not documented, LLMs cannot plan downstream calls without knowing what fields are returned. No error handling guidance is provided; failed calls offer no recovery path. Moderate quality, functional but missing several pattern implementations required for production agent use.

• Document output schema for every tool. For list_functions, explicitly state: 'Returns {items: [{addr: string (hex), name: string, size: int, n_bb: int}], total: int}'. For disasm_json, return structure: '{addr: string, count: int, ops: [{addr, mnemonic, operands}]}'. LLMs need explicit field names to compose tools.
• Add error recovery guidance to tool descriptions. E.g., disasm_by_func_text: 'If the function is >5000 instructions, returns timeout error. Try disasm_text (linear disassembly) with a smaller count instead. If addr is invalid, returns 404, use list_functions to find valid addresses.'
• Define bounds and constraints for numeric parameters. E.g., 'limit: 1-1000 (default 50). Requesting >1000 items causes timeout.' Add regex pattern for addr: '^0x[0-9a-fA-F]+$' in schema. Define min_length bounds: 1-10000.
• Rewrite descriptions to include WHEN and WHY to use each tool. E.g., 'Use disasm_by_func_json for small functions (<500 instructions) that need precise operand parsing. For large functions or when you only need mnemonics, use disasm_text. If decompilation is available, prefer decompile for human-readable pseudocode.'
• Document pagination response format in list_* descriptions. State: 'Returns {items: [...], total: int, offset: int}. If total > offset + limit, more items exist, call again with offset=offset+limit to fetch next page.'
• Clarify write tool behavior: rename_function is idempotent (safe to retry with same addr and new_name). set_comment appends or overwrites, clarify which. All write tools return {status: 'ok'} on success or raise an exception (specify which exceptions) on failure.