Loading ToolBench
The full-stack MCP that bridges frontend and backend analysis — sees what single-side tools miss.
CodeMax has 5 tools with clear, descriptive names and good descriptions (avg 180 chars, within baseline 194). All tools have input schemas with type definitions and descriptions. However, output schemas are NOT documented, responses are inferred from code (JSON report, text summary, structured health object). Parameter descriptions are adequate but lack format constraints (e.g., projectPath format, issue description length). No enum constraints, no pagination, no error recovery guidance. Tools are READ_ONLY but lack explicit idempotentHint annotations. Composition is sound (each tool does one thing), but output chaining is weak, tools return different structures, making downstream composition difficult.
Compare frontend API calls against backend routes. Find phantom calls (frontend → nowhere), dead endpoints (backend → unused), method mismatches, auth gaps, and field drift.
Run a comprehensive full-stack analysis — scans frontend API calls, backend routes, cross-references contracts, detects mismatches, auth gaps, dead endpoints, and scores overall health. The complete picture.
Quick full-stack health score with the top 5 issues. Faster than a full audit — use for a pulse check.
Map all connections between frontend components and backend endpoints. Shows which frontend files depend on which API routes, plus orphaned endpoints and phantom calls.
Given a bug description or error message, determine whether it's a frontend, backend, or cross-stack issue. Traces through the call chain and identifies the root cause.
Output schemas not documented. Tools return JSON/text but LLMs cannot infer field structure for downstream chaining. health_check returns {health: {grade, overall, dimensions}, topIssues: []}, but this is inferred from code, not declared.
No error recovery guidance. Error responses are generic ('Error running full-stack audit: <message>'). LLMs cannot determine if error is retryable, user-fixable, or fatal. No suggestions for next steps.
Parameter constraints missing. projectPath accepts any string but should validate it is an absolute path and exists. issue parameter in trace_issue has no length limit or format guidance. No enum constraints anywhere.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | <=2025-11-25 | v2 |
Tool annotations missing. All tools are READ_ONLY but lack explicit idempotentHint or readOnlyHint annotations in schema. LLMs cannot infer safety properties without explicit hints.
No pagination or result limits documented. full_stack_audit and map_dependencies may return large reports. No limit parameter, no pagination guidance, no token cost warning for LLMs.