HTTP server plugin for Ghidra that exposes program analysis data and reverse engineering capabilities through an MCP interface
GhidraMCP provides 27 tools for binary reverse engineering with basic naming and descriptions, but has critical quality gaps. All tools have name-verb structure (positive: list_*, get_*, decompile_*, rename_*, search_*, set_*) and non-empty descriptions. However, descriptions are generic (average ~60 chars, well below production baseline of 194 chars) and lack crucial context: no explanation of WHEN to use tools, no indication of which tools are mutually exclusive, no guidance on multi-step workflows. Schemas are visible for all tools (offset/limit pagination params are properly typed), but parameter descriptions are minimal (10-30 chars each). Critical gap: tool descriptions do not declare whether operations are read-only vs write, agents cannot determine safety/idempotence from tool name or description alone. Error handling is absent from all tool definitions. Output schemas are undocumented, agents must guess what fields decompile_function or get_xrefs_to return. No dependency hints (e.g., 'search_functions_by_name returns only names; call decompile_function_by_address to get code'). Composition is weak: redundant tools (rename_function vs rename_function_by_address, decompile_function vs decompile_function_by_address) create LLM decision fatigue without clear naming distinction. Per-tool average ~42 reflects consistent mediocrity across all 27 tools rather than a few standout examples.
Decompile a specific function by name and return the decompiled C code.
Decompile a function at the given address.
Get assembly code (address: instruction; comment) for a function.
Get the address currently selected by the user.
Get the function currently selected by the user.
Get a function by its address.
Get all references to the specified function by name.
Descriptions lack critical operational context: no WHEN to use each tool, no indication of read-only vs mutating operations, no workflow guidance. Agents cannot determine if a tool is safe to retry or has irreversible side effects.
Output schemas are completely undocumented. Agents must guess what fields decompile_function, get_xrefs_to, list_strings, and discovery tools return. No chaining hints.
Redundant tool pairs without clear naming distinction create decision fatigue for LLMs: rename_function vs rename_function_by_address, decompile_function vs decompile_function_by_address. Either consolidate or use verb modifiers (e.g., rename_by_name / rename_by_address).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 53 | 2026-07-28+ | v2 |
| 2026-03-09 | D | 55 | - | v1 |
Get all references from the specified address (xref from).
Get all references to the specified address (xref to).
List all namespace/class names in the program with pagination.
List defined data labels and their values with pagination.
List exported functions/symbols with pagination.
List all functions in the database.
List imported symbols in the program with pagination.
List all function names in the program with pagination.
List all non-global namespaces in the program with pagination.
List all memory segments in the program with pagination.
List all defined strings in the program with their addresses.
Rename a data label at the specified address.
Rename a function by its current name to a new user-defined name.
Rename a function by its address.
Rename a local variable within a function.
Search for functions whose name contains the given substring.
Set a comment for a given address in the function pseudocode.
Set a comment for a given address in the function disassembly.
Set a function's prototype.
Set a local variable's type.
Parameter descriptions are minimal (10 - 30 characters, vs 72 char production baseline). Examples: 'Memory address in hex format' (28 chars) lacks guidance on format ('0x1400010a0' vs '1400010a0' vs '140010a0'). 'Current function name' is vague, ambiguous vs display name or internal symbol.
No error handling guidance. If rename_function or set_function_prototype fails, agents receive no actionable next step (e.g., 'Function not found, try search_functions_by_name first').
get_current_address and get_current_function have empty input schemas. While this is not incorrect (no params needed), descriptions should clarify dependency: 'Returns the address/function the user has selected in Ghidra UI. Requires Ghidra GUI to be active.' to manage expectations.
Tool descriptions average ~60 characters (baseline 194 chars). Most lack WHEN to use, preconditions, or downstream implications. Example: 'List all functions in the database.' (37 chars), but is pagination supported? How many functions exist? When should agents call this vs search_functions_by_name?
No indication of resource constraints or limits beyond pagination defaults (offset/limit). Tools like list_strings default to 2000 items, agents may not realize this is a hard cap and could timeout or exhaust context.