Bright Data Web MCP Skill
Reliable web access for MCP-compatible agents: search, scrape, extract structured data, and automate browsers with automatic anti-bot and CAPTCHA bypass.
Docs disclose least-privilege tool scoping (groups/tools params), include security and compliance sections (treat scraped content as untrusted, respect robots.txt, avoid personal data), and clearly document data flow (API token, SSE/HTTP endpoints). Deducted for: no rollback/confirmation mechanisms; API token embedded in URL query strings (leakage into logs); no control over Bright Data's commercial service and 60+ tool behaviors; tool descriptions could act as instructions.
Internally consistent, well-structured docs with troubleshooting (npx ENOENT, timeouts, auth) and a verify-setup step; static review without execution evidence, so capped; no tests, thin coverage of error-response formats and abnormal-input behavior, failure feedback depends on the upstream service.
Clear trigger conditions (description names live web content, CAPTCHA bypass, e-commerce/social data), scenario tables and tool-selection guidance, reasonable 'when to use' boundaries. Deducted for: core function depends entirely on the overseas mcp.brightdata.com endpoint with no mainland-China reachability statement; no Chinese-language support; Pro mode is paid and free-tier boundaries only partially explained.
Good layered docs (SKILL.md plus five reference files), progressive disclosure, install notes, parameter tables, troubleshooting, and clear repo-level MIT license. Deducted for: no per-skill versioning or changelog, no declared maintenance ownership or update path, and claims like '60+ tools' may drift with upstream releases.
Claims cover search, scraping and structured extraction with rich examples and defined output formats, but static review yields no executed verification, capped at 7. Deducted for: all outputs depend on a third-party paid/quota service — unusable without a Bright Data account; marginal value vs. manual methods or alternatives (e.g. Firecrawl) not evidenced.
Auditable primary sources referenced (Bright Data's official GitHub, docs, Smithery) plus community demos, but these are external claims; no committed test suites or CI covering this skill's key paths; figures like '30-60% token savings' are independently unreproduced, hence the notable deduction.
- This skill is a wrapper around Bright Data's commercial MCP service; registration and an API token are required — it is entirely unusable without an account.
- Remote endpoints place the API token in URL query strings, risking leakage into logs, history, or proxies; prefer local mode with environment variables.
- Core function depends on the overseas mcp.brightdata.com endpoint; mainland-China reachability is unverified — Chinese users should test connectivity themselves.
- Pro mode and most structured tools are paid; the free tier (5,000 requests/month) only covers basic search and scraping.
- Skill content is static documentation with no committed tests or CI verification; capability claims like '60+ tools' may drift with upstream service versions.
- Scraping raises ToS and compliance risks with target sites; CAPTCHA/anti-bot bypass may be legally contested in some jurisdictions — assess before use.
What does this skill do, and when should you use it?
This skill wraps Bright Data's Web MCP service, letting agents search the web, fetch any URL as clean Markdown or HTML, extract structured JSON via AI prompts, and drive a full browser when needed. It handles anti-bot measures and CAPTCHAs automatically and ships prebuilt structured-data extractors for Amazon, eBay, LinkedIn, Instagram, and many other platforms. The free Rapid mode includes 5,000 requests per month; Pro mode unlocks 60+ advanced tools with selective tool-group subscriptions. Setup supports a remote SSE/HTTP endpoint requiring no installation, or a local npx-based run.
Calls search_engine to fetch Google/Bing/Yandex SERP results (JSON for Google, Markdown for the others); uses scrape_as_markdown to convert pages to clean Markdown with anti-bot bypass; runs extract for AI-powered JSON extraction from custom prompts; parallelizes up to 10 requests via search_engine_batch and scrape_batch; drives browser sessions with scraping_browser_* tools for navigation, clicks, typing, screenshots, and network-request inspection; and pulls structured data through web_data_* extractors for e-commerce, social media, finance, and business platforms.
- Developers building research agents: fetch live web content when standard HTTP requests are blocked by anti-bot measures or CAPTCHAs, using search plus scraping.
- Users doing e-commerce price comparison or market analysis: pull structured product names, prices, and reviews in bulk from Amazon, eBay, Walmart, and similar platforms.
- People monitoring social media: extract structured data from LinkedIn posts, Instagram profiles, TikTok content, and more.
- Users needing JavaScript-rendered sites: interact with dynamically loaded content via clicks, scrolls, and typing using browser automation.
- Engineers building batch data pipelines: process up to 10 URLs or queries per request with scrape_batch and search_engine_batch.
What are this skill's strengths and limitations?
- Free Rapid mode with 5,000 requests/month, usable via remote endpoint with no installation
- Automatic anti-bot and CAPTCHA handling, including dynamic and JS-heavy sites
- Configurable at tool granularity: subscribe by group (9 groups including ecommerce, social, finance) or individual tools instead of loading all 60+ Pro tools
- Prebuilt structured extractors for Amazon, LinkedIn, Instagram, YouTube, and other major platforms — faster and more reliable than raw scraping
- SKILL.md includes thorough troubleshooting (spawn npx ENOENT, timeout tuning) and workflow patterns — high documentation quality
- Depends on the third-party commercial Bright Data service; requires an API token, and Pro features plus most web_data_* tools are paid
- While the repo is MIT-licensed, the skill is a service wrapper — actual use is governed by Bright Data's terms of service
- 180s timeouts recommended for complex sites and browser automation operations must be kept close together, or sessions may be flaky
- No automated test suite or independent benchmarks in the source; reliability claims come from the documentation itself
How do you install this skill?
Option 1 (recommended, remote): configure the SSE endpoint https://mcp.brightdata.com/sse?token=YOUR_API_TOKEN or the Streamable HTTP endpoint https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN in your MCP client — no local installation needed. Option 2 (local): run API_TOKEN=<token> npx @brightdata/mcp, which requires Node.js. Both require a Bright Data API token. The skill lives at hugging-face-skills/skills/brightdata-web-mcp/ in patchy631/ai-engineering-hub; how to place it into a specific client's skills directory is not documented in the source.
How do you use this skill?
Once configured, invoke tools on demand from any MCP-capable client. A typical research flow: call search_engine (e.g. {"query": "latest AI news", "engine": "google"}) to find URLs, then scrape_as_markdown for content, then extract with a custom prompt for structured JSON. Enable Pro mode with &pro=1 on the remote URL or PRO_MODE=true locally, or subscribe selectively via GROUPS (e.g. ecommerce,social) or the TOOLS parameter.
How does this skill compare with similar options?
The repo's MCP project collection (e.g. MCP Agentic RAG Firecrawl, Cursor Linkup MCP) demonstrates similar web-scraping MCPs like Firecrawl and Linkup that could serve as alternatives; however, the source provides no direct comparison data.