Local API simulation and test evidence for AI-assisted development. Provides mock API responses, traffic logging, fault injection, and MCP tools for testing AI agents.
GhostAPI provides 4 tools with explicit schemas and descriptions. However, quality is uneven. Tool names lack clear action verbs (inspect_state, toggle_chaos_mode are vague). Descriptions are present but brief (avg ~60 chars, below 194-char baseline). Parameter descriptions exist but lack depth, no guidance on when to use each tool, dependencies between them, or recovery paths. Schema completeness varies: set_api_behavior has detailed constraints (HTTP methods enum, status range 100-599, delay bounds); toggle_chaos_mode uses union types but lacks clarity on mutual exclusivity. Output schemas are undocumented, LLMs cannot predict what inspect_state or get_traffic_logs return. Error handling is absent, no guidance on what happens if state.json is missing, if chaos mode conflicts with active requests, or how to recover from invalid configurations.
Return the latest 50 captured HTTP requests from the in-memory dashboard buffer.
Return the current local object state from .ghostapi/state.json.
Configure a deterministic local response for a method/path pair.
Enable or disable local latency/errors for resilience testing.
Tool names lack action verbs. 'inspect_state' should be 'get_state'; 'toggle_chaos_mode' should be 'enable_chaos_mode' or 'set_chaos_mode'. LLMs infer intent from verb-noun patterns, vague names reduce tool discoverability.
Output schemas undocumented. inspect_state and get_traffic_logs lack return type definitions. LLMs cannot plan downstream operations or extract required fields without knowing response structure.
No error handling guidance. If .ghostapi/state.json is missing, if chaos mode conflicts with active requests, or if invalid HTTP methods are passed, tools provide no recovery hints. Errors should guide LLMs to next steps.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2026-07-28+ | v2 |
Parameter descriptions lack depth. 'Response body' and 'Response delay in milliseconds' are minimal. Descriptions should explain WHEN to use each parameter, constraints, and dependencies (e.g., 'delayMs simulates network latency; max 10s to prevent test timeouts').
No tool composition guidance. Unclear whether set_api_behavior and toggle_chaos_mode can be used together, or if chaos mode overrides configured behaviors. Dependencies between tools should be documented.