Loading ToolBench
Atomic MCP tools for AppSec and AI Security engineers
14 tools with highly specialized security audit/testing functions. Most tools lack visible input schemas in the provided source (only 2 of 14 have explicit schema definitions shown: jwt_inspect and agent_tool_risk_audit). Descriptions are present but vary widely in quality, some are detailed (agent_tool_risk_audit: 280+ chars), others minimal (owasp_llm_classify, sensitive_files_list: <50 chars). No output schemas documented. No error handling guidance visible. Tool names are action-oriented (audit, inspect, generate, register, poll, stop) but several are domain-specific jargon that may not clearly signal intent to a general LLM (phpggc_generate, interactsh_register). Security-focused toolkit with legitimate red-team use cases, but definition quality lags production baselines.
Statically audit a single agent tool definition for schema-level risks. Accepts OpenAI function-calling, Anthropic tool-use, MCP tool, or a bare JSON Schema. Reports: over-broad params (bare-string paths/commands/URLs), missing constraints (enum/pattern/min/max/maxLength/maxItems), dangerous defaults (suspicious paths, disabled safeguards), exfil-shape (URL-destination + data-payload in the same tool), ambiguous descriptions vs risky tool names.
Look up default credentials for common services and devices
Introspect a GraphQL endpoint and extract schema information
Compare two HTTP responses and report differences
Poll Interactsh for recorded interactions
Register a new Interactsh interaction server for out-of-band testing
Stop an Interactsh interaction server
12 of 14 tools lack visible input schemas in source code. Only agent_tool_risk_audit and jwt_inspect show explicit parameter definitions.
9 tools have descriptions under 50 characters (prompt_injection_audit, owasp_llm_classify, http_diff, default_creds_lookup, sensitive_files_list, graphql_introspect, phpggc_generate, interactsh_poll, interactsh_stop).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 33 | 2026-07-28+ | v2 |
Inspect and validate JWT tokens, check for weak algorithms and secrets
Heuristic static analyzer for MCP server source files. Parses a Python file that defines a FastMCP (or low-level Server) and reports on each registered tool: parameters, docstring quality, and a set of risk heuristics (shell exec, filesystem writes, network egress, over-broad params).
Classify a security finding or vulnerability against OWASP LLM Top 10
Generate PHP gadget chain payloads for deserialization attacks
Audit a prompt/instruction string for injection patterns and risky markers
Generate a list of sensitive files commonly found in systems
Generate wordlists for password cracking and fuzzing
No output schemas documented for any tool. LLMs cannot infer what fields to expect or plan downstream calls. Baseline: 100% of A+ tools document return types.
Tool names phpggc_generate and interactsh_* use domain-specific jargon (PHPGGC, Interactsh) that may not clearly signal intent to general LLMs. Consider: generate_php_gadget_chain, register_oob_server, poll_oob_interactions.
No error handling guidance visible in any tool description. Per pattern:recovery-guide, error responses should tell LLMs what to do next (e.g., 'Try search_users() with a partial name'). None present.