Loading ToolBench
Local-first eval, runtime control, and audit replay for AI agents. Open-source security testing for AI agents with red-team evals, report compare, local guard, and runtime policy.
AntiClaude is a security red-teaming CLI tool, not a traditional MCP server. Tool definitions are inferred from file paths and command names rather than explicitly registered via MCP protocol. Only 1 of 9 tools (scan) has a visible input schema with types and descriptions. The remaining 8 tools lack documented schemas entirely. Descriptions are present but generic (10-50 chars). No parameter descriptions visible for most tools. No output schemas documented. This is a CLI application masquerading as an MCP server, it does not implement the MCP protocol for tool registration, schema validation, or structured communication.
Audit AI agent skills and capabilities for security vulnerabilities
Generate security badge for CI/CD pipelines
Compare two scan reports to detect regressions and new vulnerabilities
Load and manage test fixtures for mock agents
Evaluate runtime guard policies against tool calls and agent outputs
Discover and audit MCP (Model Context Protocol) servers for security issues
Replay and audit historical agent execution traces
Manage review queue for flagged agent decisions
No MCP protocol implementation detected. This is a CLI tool, not an MCP server. Tools are not registered via MCP's tool definition mechanism; they are inferred from file paths and command names.
8 of 9 tools lack input schemas entirely. Only 'scan' has a documented schema. Without schemas, LLMs cannot validate inputs or understand parameter types and constraints.
Tool descriptions are generic and under 50 characters (e.g., 'Audit AI agent skills and capabilities for security vulnerabilities'). No guidance on WHEN to use each tool, WHAT it returns, or HOW it differs from similar tools.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 29 | 2026-07-28+ | v2 |
Run security scan against an AI agent endpoint with configurable payloads and adapters
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields without knowing what each tool returns.
No error handling guidance visible. Tools do not document what errors can occur, whether they are retryable, or what the LLM should do next.