AI-first web browser on NW.js with zero CORS, full Node.js integration, API-steerable control, and MCP-enabled tool support
Nightmare Browser exposes 27 tools with basic schemas and minimal descriptions. Most tools have 1-2 sentence descriptions (avg ~40 chars, well below the 194-char baseline for A+ tools). Parameter descriptions are entirely absent, only type and required status are declared. No output schemas documented. Tools lack error recovery guidance. Many tools in trash/ are not integrated into the active codebase. The server follows a verb_noun naming convention (nightmare_*) but descriptions are too terse to guide LLM selection. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk classifications in the metadata.
Activate a tab by ID
Add a bookmark
Click element by CSS selector
Close a tab by ID
Open a new tab with optional URL
Duplicate a tab
Execute JavaScript in tab context
Parameter descriptions entirely absent. All 27 tools declare parameter types and required status but provide zero guidance on what each parameter controls, valid ranges, or expected formats. LLMs cannot infer intent from bare type names like 'selector', 'level', 'query', or 'timeout'.
Tool descriptions are too terse (avg ~40 chars vs 194-char baseline). Descriptions like 'Open a new tab with optional URL' and 'List all open tabs with state' lack context on when to use each tool, what they return, and how they differ from similar tools. LLMs cannot distinguish nightmare_get_tab from nightmare_get_state without richer descriptions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
Find text in current page
Read console log buffer for a tab
Get page HTML content
Full browser state snapshot
Get full tab state by ID
Navigate back in tab history
Navigate forward in tab history
List all open tabs with state
Toggle mute state on a tab
Navigate a tab to a URL
Toggle pin state on a tab
Query DOM element by selector
Relaunch browser without killing process
Reload a tab
Take screenshot of tab
Search browsing history
Shutdown the browser
Type text into element by selector
Wait for selector or condition
Set zoom level for a tab
No output schemas documented. Tools return structured data (e.g., nightmare_get_tab returns tab state, nightmare_screenshot returns image data) but the response format is not declared. LLMs cannot plan downstream tool calls or extract required fields without knowing what fields are returned.
No tool annotations despite clear risk classifications. Tools are marked DESTRUCTIVE (nightmare_close_tab, nightmare_shutdown) or WRITE (nightmare_navigate, nightmare_execute_js) in metadata but lack MCP tool annotations (readOnlyHint, destructiveHint, idempotentHint). Agents cannot distinguish safe reads from irreversible operations.
No error recovery guidance. Tools throw generic errors ('Tab not found', 'Unknown tool') without suggesting next steps. Per pattern:recovery-guide, errors should tell LLMs what to do: 'Tab not found. Call nightmare_list_tabs() to see available tabs.'