A fast web retrieval engine with HTTP-to-browser fallback, producing LLM-ready Markdown. Provides MCP tools for web scraping, crawling, URL discovery, and data extraction.
Essence provides 5 well-named, verb-prefixed tools (scrape, crawl, crawl_stream, map, extract) with comprehensive input schemas and clear descriptions. All tools have 50-150 character descriptions explaining purpose and use case. Input parameters are typed and described (e.g., 'url', 'max_depth', 'engine' with enum values). However, output schemas are not documented in the source, the code shows request types but no explicit response schema definitions. Error handling is present (ScrapeError enum, timeout detection) but lacks recovery guidance in descriptions. Parameter relationships (e.g., 'mode' in extract determines which of 'selectors' vs 'schema' apply) are underdocumented. No tool annotations (readOnlyHint, destructiveHint) visible. All tools are READ_ONLY by risk classification, but this is not declared in the tool definitions themselves.
Crawl a website starting from a URL, discovering and scraping multiple pages up to a specified depth and limit.
Stream crawl events as Server-Sent Events (SSE). Streams status updates, completed documents, errors, and final summary in real-time.
Extract structured data from URLs using CSS selectors, LLM-based extraction, or automatic mode. Supports multiple extraction modes and custom schemas.
Discover URLs from a given URL by fetching sitemaps and parsing HTML links. Supports filtering by subdomain, search query, and URL limits.
Scrape a URL and return content in multiple formats (markdown, rawHtml, text, json). Supports both HTTP and browser-based retrieval with automatic fallback.
Output schemas not documented. Code shows request types (MapRequest, ScrapeRequest) but response structures are inferred from implementation, not explicitly declared. LLMs cannot plan downstream tool calls without knowing what fields to expect.
Tool annotations missing. All tools are READ_ONLY (no state mutation), but this is not declared via readOnlyHint in tool definitions. Agents cannot distinguish safe-to-retry tools from destructive ones without explicit hints.
Error recovery guidance absent. ScrapeError enum exists (InvalidUrl, Timeout, RequestFailed, Internal) but tool descriptions do not explain what to do on failure. E.g., 'If timeout occurs, try increasing timeout parameter or using engine=http instead of browser.'
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | <=2025-11-25 | v2 |
Parameter dependencies underdocumented. In extract, 'mode' enum (css|llm|auto) determines which of 'selectors' vs 'schema' vs 'prompt' are valid, but this constraint is not stated in parameter descriptions.
Pagination not visible in tool definitions. crawl and map likely return large result sets, but no limit/offset/cursor parameters are documented in the input schemas provided.