NexusX Model Context Protocol Server for web scraping, crawling, content extraction, and browser automation
Server has 10 tools with complete JSON Schema definitions and descriptions. Naming follows verb_noun convention (scrape_, crawl_, extract_, search_, create_, delete_, list_). Descriptions are present and substantive (avg ~150 chars). However, several critical gaps: (1) Output schemas are NOT documented, no tool specifies what fields are returned, forcing LLMs to guess downstream structure. (2) Error handling is absent, no guidance on retryability, user-fixable vs fatal errors, or recovery paths. (3) Parameter descriptions lack actionable constraints (e.g., 'maxPages' has min/max but no guidance on typical values or performance impact). (4) sessionOptions is a complex nested object repeated across 7 tools, should be abstracted or documented with clearer dependency hints. (5) No tool annotations (readOnlyHint, destructiveHint) despite clear risk levels (READ_ONLY, WRITE, DESTRUCTIVE). Strengths: all tools have descriptions, all have input schemas with types, parameter enums are properly declared (outputFormat, profile options).
Fast, lightweight browser automation with the Browser Use agent
Complex browser tasks using Claude computer use
Crawl a website starting from a URL and explore linked pages. This tool allows systematic collection of content from multiple pages within a domain. Use this for larger data collection tasks, content indexing, or site mapping.
Creates a new persistent NexusX profile.
Deletes an existing persistent Data profile.
Extract structured data from webpages using a JSON schema and natural language prompt
Output schemas not documented. No tool specifies what fields are returned (e.g., scrape_webpage returns what structure? markdown string? object with content+metadata?). LLMs cannot plan downstream tool calls or extract required fields without guessing.
No error handling guidance. Tools lack descriptions of failure modes, retryability, or recovery paths. E.g., what happens if scrape_webpage hits a 403? Is it retryable? Should the agent try a different URL? No guidance.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
Lists all persistent NexusX profiles.
Complex browser tasks using OpenAI Computer Use API
Scrape a single webpage and extract its content in various formats
Search the web using Bing. This tool allows you to search the web using bing.com
Tool annotations missing. Tools declare risk levels (READ_ONLY, WRITE, DESTRUCTIVE) in metadata but do not use MCP readOnlyHint, destructiveHint, or idempotentHint annotations. LLMs cannot infer safety properties from names alone.
sessionOptions parameter is complex nested object repeated across 7 tools. No guidance on when to set useProxy, useStealth, solveCaptchas, or acceptCookies. Descriptions say 'Recommended false' but do not explain trade-offs (performance vs detection evasion).
Parameter constraints lack actionable context. maxPages has min=1, max=100, default=10, but no guidance on typical values, performance impact, or when to increase. numResults in search_with_bing defaults to 5 but does not explain why or when to override.