Scan an MCP tool manifest for tool-poisoning indicators
Paste MCP tool definitions (a tools/list response, a { "tools": [...] } object, a bare array, or a single tool) for a heuristic scan of name, description and inputSchema: hidden characters, look-alike identifiers,
coercive phrasing, data-egress framing, shadowed names and schema-embedded instructions.
A heuristic advisory, not a verdict. These are heuristic indicators for review — they require human judgement and are not proof of ill intent. Every result is an indicator for review. The manifest is parsed as data and never executed, analysed entirely in your browser, and never uploaded. A saved permalink stores only the derived findings with short, redacted evidence — never your raw manifest, which may contain internal tool or business names.