Loading ToolBench
Python Code Intelligence Engine — static analysis knowledge graph exposed via MCP for AI agents
Pyxus exposes 4 read-only code analysis tools via FastMCP. Tool names are action-oriented (context, impact, search, imports) and descriptions are present but lack depth. Input schemas are visible and typed, but parameter descriptions are minimal. Output is JSON-serialized without documented structure. No error recovery guidance, no parameter constraints (enums), and no tool annotations. Overall: functional but below production baseline.
Get a 360-degree view of a symbol: methods, callers, callees, imports, inheritance.
Analyze blast radius: what breaks if you change this symbol.
Analyze module-level import dependencies and detect circular imports. Returns the dependency graph between modules, circular import chains, and per-module import/imported-by counts.
Search symbols by name.
Tool definitions in minified JavaScript bundle (index-1ABEpEQp.js) cannot be verified. 'context' and 'search' tools lack visible schema and parameter documentation in source code.
No output schema documented. Tools return JSON strings without specifying field names, types, or structure. LLMs cannot plan downstream calls or extract data reliably.
Parameter descriptions are minimal or missing. 'query_str' in search() lacks guidance on format/length. 'direction' in impact() has no enum constraint, LLM must guess valid values.
No error recovery guidance. Tools return JSON error objects without actionable next steps. LLM cannot distinguish retryable vs fatal errors.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | 2026-07-28+ | v2 |
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). All tools are read-only but this is not declared in the schema, forcing LLMs to infer safety.