A collection of MCP server plugins for reverse engineering, security analysis, and CTF tools including Ghidra bridge, ReverseLab tools, and red team hardening capabilities
Two tools with basic descriptions but critical schema gaps. list_instances has a good description (194 chars) but empty input schema ({}). connect_instance has a strong description (189 chars) and a minimal schema with one string parameter, but lacks output schema documentation. Neither tool documents what fields are returned, forcing LLMs to infer structure. No parameter descriptions beyond the tool level. No error handling guidance. Naming is clear (verb_noun pattern), but schema completeness is severely lacking.
Switch the MCP bridge to a different Ghidra instance by project name. Before calling this function only the static bridge tools are exposed (list_instances, connect_instance, tool-group management, debugger proxy). After a successful connect the bridge fetches the instance's /mcp/schema and registers Ghidra analysis tools dynamically.
List known Ghidra instances from UDS discovery and the active TCP fallback. Returns JSON with each instance's project name, PID, open programs, and socket path or TCP URL. Also shows which instance is currently connected.
No output schemas documented for either tool. LLMs cannot predict return structure (fields, types, nesting). list_instances returns JSON with 'instances' array and per-instance fields (project, PID, socket, url, connected), none of this is formally declared. connect_instance returns a dict with 'connected', 'transport', 'project', 'socket', 'tools_registered', etc., also undocumented.
list_instances has empty input schema ({}). While correct for a no-argument tool, the schema should be explicitly documented as 'no parameters required' in the description to guide LLM usage.
connect_instance 'project' parameter lacks description. The parameter name alone does not explain: Is it a project name or ID? What if the project doesn't exist? What happens on mismatch? The code shows fuzzy matching and fallback logic, but none of this is exposed to the LLM.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | 2026-07-28+ | v2 |
No error handling guidance. connect_instance can fail with 'Schema fetch failed', 'No instance matching', 'Invalid TCP URL', or 'cancelled', but the tool description does not explain these failure modes or what the LLM should do next (retry, ask user, try list_instances first).
No pagination or result limits documented. list_instances returns all discovered instances without limit. If 100+ Ghidra instances are running, the response could be very large. No guidance on whether results are paginated or capped.