Loading ToolBench

Reading the current corpus…

MCP_Tool_Poisoning_Detector MCP server · F · ToolBench

← Browse

MCP_Tool_Poisoning_Detector

github.com/treasonking/MCP_Tool_Poisoning_Detector

Static security scanner for malicious natural-language instructions and permission mismatches in MCP tool metadata.

2026-07-28+2 tools

25/100

Definition Quality35

Protocol Readiness0

SupportabilityNot measured-

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence

Current-spec patterns detected

Summary

This is a static analysis tool (scanner), not an MCP server. The repository contains sample vulnerable and safe MCP servers for testing purposes, but does NOT itself expose MCP tools. The two tools listed (project_file_reader, safe_project_helper) are demo implementations in samples/, not registered MCP tools. No MCP server implementation, transport, or tool registration visible. The actual scanner is a CLI tool (mcp-toolpoison-scanner) that analyzes OTHER servers' metadata and code.

Tools (2)

project_file_readerread onlysource verified40/100

Reads a project file selected by the user.

safe_project_helperdestructivesource verified30/100

Helps with project tasks.

Issues (5)

CRITICALtool ↗

Repository is a static analysis scanner, not an MCP server. No MCP tool definitions, transport, or server implementation present.

project_file_readersafe_project_helper

CRITICALtool-description ↗

Sample tools lack input parameter descriptions. 'path' and 'command' parameters have null descriptions, violating pattern:tool-description requirement that every parameter must have a non-empty description.

project_file_readersafe_project_helper

HIGHtool ↗

Tool names are not action-verb prefixed. 'project_file_reader' should be 'read_project_file'; 'safe_project_helper' is generic and ambiguous (does not convey what action it performs).

project_file_readersafe_project_helper

HIGHtool ↗

No output schemas documented for either tool. LLMs cannot infer what fields or structure to expect from responses.

project_file_readersafe_project_helper

Recommendations

If this repository is intended to be an MCP server itself (not just a scanner), implement proper MCP server transport (HTTP/SSE or Streamable HTTP) and register tools via the MCP protocol.
Add non-empty descriptions to all input parameters. Replace null with descriptive text explaining what each parameter controls and its expected format.
Rename tools to action-verb format: 'read_project_file' instead of 'project_file_reader'; replace 'safe_project_helper' with specific action names like 'execute_project_command' (if intentional) or split into separate tools.
Document output schemas for each tool, specifying field names, types, and what data is returned. Include pagination info if results can be large.
For the vulnerable sample tool (safe_project_helper), add input validation, command whitelisting, and permission checks to prevent command injection attacks.

Score history

Overall score trend

First recorded score · v2 rubric

25/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	F	25	2026-07-28+	v2