MCP server for Firecrawl web scraping integration. Supports both cloud and self-hosted instances. Features include web scraping, batch processing, structured data extraction, and LLM-powered content analysis.
Firecrawl MCP server has well-structured tool definitions with complete input schemas and clear descriptions. All 6 tools have descriptions (194 - 250 chars, within baseline 34 - 392 range) and detailed parameter schemas with types and enums. However, output schemas are not documented, LLMs cannot predict response structure for downstream chaining. Parameter descriptions are present but some lack actionable constraints (e.g., 'limit' has no min/max bounds). Error handling and recovery guidance are absent. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all tools being read-only operations. Tool names follow verb_noun convention (firecrawl_scrape, firecrawl_map, firecrawl_crawl) but lack clarity on mutual exclusivity (e.g., firecrawl_crawl vs firecrawl_batch_scrape both do multi-page operations).
Scrape multiple URLs in batch mode. Returns a job ID that can be used to check status.
Check the status of a batch scraping job and retrieve results when complete.
Start an asynchronous crawl of multiple pages from a starting URL. Supports depth control, path filtering, and webhook notifications.
Discover URLs from a starting point. Can use both sitemap.xml and HTML link discovery.
Scrape a single webpage with advanced options for content extraction. Supports various formats including markdown, HTML, and screenshots. Can execute custom actions like clicking or scrolling before scraping.
Search for content across the web using Firecrawl's search capabilities.
Output schemas not documented. LLMs cannot predict response structure (fields, types, pagination) for downstream tool chaining or data extraction. Baseline: 100% of A+ tools document return types.
Numeric parameters lack bounds. 'limit' and 'maxDepth' have no min/max constraints, inviting LLMs to pass absurd values (e.g., limit=999999). Baseline: constrained-input pattern requires enums or range declarations.
No error handling guidance. Tool descriptions do not explain failure modes, recovery steps, or when to retry. Baseline: error-classification pattern requires categorizing errors as retryable, user-fixable, or fatal.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
No tool annotations. All tools are read-only (safe to retry) but lack readOnlyHint annotation. Batch operations (firecrawl_batch_scrape) should declare idempotentHint. Annotations guide agent retry logic.
Semantic overlap: firecrawl_crawl and firecrawl_batch_scrape both scrape multiple pages. Descriptions do not clarify when to use each (crawl is async/depth-based; batch is URL-list-based). LLMs conflate similar names.