MCP server for React Native runtime debugging — zero-config access to Redux state, AsyncStorage, actions, logs, network, and profiling from your running app
This MCP server implements 25 React Native debugging tools with solid naming conventions and reasonable schema coverage. However, there are significant gaps in consistency across tools: output schemas are not documented in any visible tool definition, many tool descriptions lack context about when/why to use them, and some parameter descriptions are minimal. The server demonstrates good intent around tool composition (one tool per concern) and parameter typing, but falls short of production-grade documentation standards. Most tools have adequate names starting with verbs (get_, take_, dispatch_, clear_, wait_for_, list_, select_), but descriptions range from sparse to adequate, and return types are entirely undocumented.
Clear all captured logs, errors, network requests, and other buffers
Dispatch a Redux action in the app
Evaluate arbitrary JavaScript code in the React Native app's runtime
Force garbage collection in the React Native app
Get Redux/state action dispatch log from the app
Get application state from Redux/Zustand stores. Zero-config: discovers Redux stores automatically via the injected runtime agent. The optional SDK adds Zustand and custom stores.
Get console logs captured from the running React Native app
Capture a CPU profile from the React Native app
Output schemas are not documented for any tool. LLMs cannot determine what fields to expect in responses, forcing them to make assumptions about available data or call patterns. This violates the tool definition contract and increases hallucination risk.
Many tool descriptions are generic or minimal (e.g., 'Get runtime errors and exceptions from the React Native app', 'Get console warnings from the React Native app'). These lack guidance on when to use each tool, what to expect, or how they differ from related tools. Average description length is ~70 chars; best practice is 100 - 200 for LLM reasoning.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 52 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 49 | - | v1 |
Get runtime errors and exceptions from the React Native app
Get failed network requests from the React Native app
Get memory usage statistics from the React Native app
Get the current navigation state and stack from the React Native app
Get navigation timing information from the app
Get network requests made by the React Native app
Get React component render profiling data from the app
Get the difference in state between two points in time
Get keys from AsyncStorage or MMKV storage
Get a value from AsyncStorage or MMKV storage
Get console warnings from the React Native app
Check the health and connection status of the React Native dev tools server
List available React Native debugging targets
Resolve a source map location in the React Native app
Select a React Native target for debugging
Capture a heap snapshot from the React Native app's JavaScript runtime
Wait for a log message matching a pattern to appear
Pagination and result limits are not documented or enforced. Tools like get_console_logs, get_network_requests, and get_action_log accept a 'count' parameter but there is no indication of maximum limits, default values, or whether pagination cursors are supported. Without this, agents may request thousands of items, exhausting context windows.
No error handling guidance is visible in tool descriptions. Tools do not document what errors can occur (e.g., 'target not connected', 'invalid store name', 'timeout'), what they mean, or what the agent should do next. This violates error classification and recovery patterns.
Destructive operations (evaluate_js, force_gc, clear_buffers, dispatch_action, select_target) lack confirmation/dry-run guidance. These tools modify runtime state and should support a confirmation or preview step to prevent accidental side effects when agents make mistakes.
Parameter 'count' in multiple tools (get_console_logs, get_errors, get_warnings, get_network_requests, get_failed_requests, get_action_log) lacks minimum/maximum constraints. Description says 'Number of recent logs to retrieve' but does not specify valid range. Without bounds, agents may pass zero, negative, or excessively large values.
Tool descriptions for profiling tools (take_heap_snapshot, get_cpu_profile, get_render_profile) do not explain what to do with the output (e.g., 'Returns a binary snapshot file that can be analyzed with Chrome DevTools') or how long the operation takes, which could help agents reason about performance impact.
Parameter descriptions lack specificity. 'Filter by log level (log, warn, error, info, debug)' in get_console_logs is clear, but most parameter descriptions are one-liners that do not explain valid values, defaults, or dependencies with other parameters.
Enum constraints are missing where appropriate. Parameter 'backend' in get_storage_keys and get_storage_value should declare an enum ('AsyncStorage' | 'MMKV') rather than relying on free-form strings. This invites hallucinated values.
No documentation of tool composition or prerequisites. For example, select_target should mention that list_targets should be called first to discover available targets. get_app_state's optional 'name' parameter is only useful after list_targets or similar discovery.