v2.9.3

v2.9.3

• New code tool group - your coding agent's companion
  • web_data_npm_package - look up any npm package by name and get back
    structured metadata: latest version, README, dependencies, and more
    (e.g. @brightdata/sdk, express, fastify)
  • web_data_pypi_package - look up any PyPI package by name and get back
    structured metadata: latest version, README, dependencies, and more
    (e.g. langchain-brightdata, requests, numpy)
  • Enable with GROUPS="code" - works with Claude Code, Cursor, Windsurf,
    and any MCP-powered coding agent
• Added code group row to the tool groups table in README
• Added coding agent configuration example in README

Why this matters

Coding agents constantly need to answer questions like "what's the latest
version of X?" or "what does this package do?". Until now that meant
scraping registry pages or relying on stale training data. The code group
gives agents a single, reliable tool to query npm and PyPI directly —
structured data, no scraping, no blocking, always up to date.

v2.8.6

What's New

scrape_batch now returns minified markdown, consistent with scrape_as_markdown.
Decorative markdown formatting is stripped from each scraped page while preserving links, code blocks, and inline code - reducing token usage when processing multiple URLs in a single call.

v2.8.5

v2.8.5 - Geo-targeted Search

Geo-targeted SERP results

Both search_engine and search_engine_batch now accept an optional geo_location parameter - a 2-letter country code (e.g. "us", "uk", "de") that lets you retrieve Google search results as seen from a specific country.

Example: Search for "best laptops" as seen from Germany:

{ "query": "best laptops", "engine": "google", "geo_location": "de" }

v2.8.3

Release Notes - v2.8.3

Release Date: January 18, 2026

New Features

Configurable Polling Timeout

• Introduced POLLING_TIMEOUT environment variable for dataset collection timeout configuration
• Default timeout: 600 seconds (10 minutes)
• Allows users to customize timeout based on their specific needs
• Applies to all web data collection tools

Browser Tools Country Targeting

• Added optional country targeting support to browser automation tools
• Enables geo-specific browser sessions for testing localized content
• Configurable via country parameter in browser tool calls

Improvements

Batch Operation Optimizations

• Reduced batch size limits from 10 to 5 for search_engine_batch and scrape_batch tools
• Updated tool descriptions to reflect new batch size constraints
• Improved reliability and performance of batch operations
• Changed batch operations to use Promise.allSettled() for better error handling

Polling Status Fixes

• Fixed polling status handling for dataset collection tools
• Improved error handling during data collection polling
• Better reliability when waiting for dataset snapshots

MCPB Bundle Update

• Updated MCPB bundle to v2.7.1
• Fixed handling of optional environment variables that could crash the server when unset
• Improved server stability and configuration handling

Configuration

New Environment Variables

# Optional: Configure polling timeout (default: 600 seconds)
POLLING_TIMEOUT=600

v2.8.1

Release Notes - v2.8.1

New Features

Tool Annotations for Enhanced LLM Understanding (#97)

• Added comprehensive tool annotations to improve how LLMs interpret and use Bright Data tools
• Enhanced tool descriptions and parameter metadata across browser and server tools
• Improved tool understanding for better autonomous agent performance

Browser Form Filling Tool (#96)

• Introduced scraping_browser_fill_form tool for efficient automated form filling
• Enables streamlined form interaction in browser automation workflows
• Added to browser tool group for enhanced web automation capabilities

Country Targeting Support (#92)

• Added optional country parameter to browser tools for geo-targeted scraping
• Enables region-specific data collection and testing
• Seamlessly integrated with existing browser tool interfaces

CI/CD Improvements

Test Workflow (#79)

• Added automated test workflow for continuous integration
• Includes server health tests
• Ensures code quality and stability before releases

v2.8.0

Release Notes - v2.8.0 🚀

Token Optimization

This release focuses on reducing token consumption for AI agents, making interactions more efficient and cost-effective.

Changes

Search Engine Improvements

• Switched web search results to parsed_light format for leaner responses
• Cleaned up search payloads to return only essential fields (link, title, description)
• Removed unnecessary fields like images, pagination, related, and ai_overview from search results

Markdown Minification

• Added markdown stripping for scrape_as_markdown responses using remark and strip-markdown
• Preserves essential elements (links, code blocks) while removing formatting overhead
• Achieves 30-60% reduction in tool response tokens

Data Sanitization

• Removed null values from dataset API responses to reduce payload size

New Dependencies

• remark: ^15.0.1
• strip-markdown: ^6.0.0

---

MCP Evaluations Framework

We've added a comprehensive evaluations suite powered by mcpjam to validate agent behavior before production.

Structure

mcp-evals/
├── server-configs/           # Server connection configs per tool group
├── tool-groups.json/         # Test cases per tool group
└── llms.json                 # LLM provider API keys

Coverage

Eval suites for all 8 Tool Groups:

• ecommerce - Amazon, Walmart, Best Buy product searches
• social - TikTok, Instagram, Twitter content
• business - LinkedIn profiles, Crunchbase, Google Maps
• research - GitHub repos, Reuters news, academic sources
• finance - Stock data, market trends
• app_stores - iOS App Store, Google Play
• browser - Scraping Browser automation
• advanced_scraping - Batch operations, custom scraping

Quick Start

# Install mcpjam
npm install -g @mcpjam/cli

# Run e-commerce evals
mcpjam evals run \
  -t mcp-evals/tool-groups.json/tool-groups.ecommerce.json \
  -e mcp-evals/server-configs/server-config.ecommerce.json \
  -l mcp-evals/llms.json

---

Performance Impact

Combined with Tool Groups and Custom Tools from previous releases:

Configuration	Total Tokens	Cost Reduction
Full Suite (No Optimization)	37,500	-
+ Tool Groups	28,500	24%
+ Custom Tools	24,750	34%
+ Token Optimization (v2.8.0)	14,500	61%

---

Resources

• Full Launch Week Documentation
• mcpjam CLI - Official MCP evaluation framework
• Bright Data Control Panel - Access Observability Dashboard

v2.7.0

What's New

Tool Groups System

Introducing flexible tool selection with curated groups! Instead of choosing between Rapid (limited) and Pro (everything), you can now:

• Mix and match tool groups using GROUPS environment variable
• Add individual tools via the TOOLS variable
• 9 curated groups: ecommerce, social, browser, finance, business, research, app_stores, travel, advanced_scraping

Example configuration:

{
  "env": {
    "GROUPS": "browser,ecommerce",
    "TOOLS": "scrape_as_markdown"
  }
}

All Changes

Features

• Add tool groups system with 9 curated categories (c23af1b)

🚀 Upgrade Guide

No breaking changes! Existing configurations continue to work:

• PRO_MODE=true still enables all tools
• Default Rapid mode unchanged
• New GROUPS/TOOLS variables are optional

v2.6.2

Bright Data MCP Server - Release 2.6.2

Release Date: Dec 1, 2025

• Added DOM-based interactive element capture with stable data-fastmcp-ref refs, improving click/scroll/type tooling on custom clickable elements not present in the filtered ARIA snapshot.
• Extended snapshot output to include a compact “DOM Interactive Elements” section when available.
• scraping_browser_snapshot now accepts filtered (default: false) to choose between raw and compacted snapshots.

v2.6.1

Bright Data MCP Server - Release 2.6.1

Release Date: Nov 16, 2025

• Pinned @modelcontextprotocol/sdk to 1.21.2 to avoid the completions capability mismatch that caused “Server does not support completions” on fresh installs.

• No other changes.

v2.6.0

Bright Data MCP Server - Release 2.6.0

Release Date: October 27, 2025

We're thrilled to announce the release of Bright Data MCP Server version 2.6.0! This major update brings significant improvements to browser automation, enhanced observability, and better resource management. Whether you're building AI agents for web scraping, search, or data extraction, this release makes your workflows more reliable and efficient.

---

Highlights

Browser Automation with ARIA References

Say goodbye to fragile CSS selectors! This release introduces a game-changing approach to browser automation using ARIA snapshots and ref-based element targeting. Your browser automation scripts are now more reliable, maintainable, and resistant to website changes.

Multi-Client Session Isolation

Working with multiple AI clients or concurrent sessions? We've implemented robust session isolation to ensure browser sessions never interfere with each other, providing clean separation and preventing cross-contamination.

---

🚀 What's New

Browser Automation Enhancements

ARIA Snapshot-Based Element Targeting

The Problem: Traditional CSS selectors break when websites update their layouts or class names, making browser automation fragile and maintenance-heavy.

The Solution: We've introduced a revolutionary ARIA snapshot system that provides stable, semantic element references:

• scraping_browser_snapshot - New tool that captures all interactive elements with stable ref identifiers
• Ref-based tools - All interaction tools now use reliable refs instead of brittle selectors:
  • scraping_browser_click_ref - Click elements by ref
  • scraping_browser_type_ref - Type into inputs by ref
  • scraping_browser_scroll_to_ref - Scroll to elements by ref
  • scraping_browser_wait_for_ref - Wait for elements by ref

Network Request Tracking

New tool: scraping_browser_network_requests

Monitor all HTTP requests made by the page - perfect for:

• Debugging API calls and data fetching
• Understanding page behavior
• Tracking AJAX requests
• Analyzing third-party integrations

See every request with method, URL, status code, and response details.

ARIA Snapshot Filtering

To reduce token consumption and improve AI response times, we've added intelligent filtering that:

• Extracts only interactive elements (buttons, links, inputs, etc.)
• Compresses snapshot data by ~70% on average
• Maintains all critical information for automation
• Filters out decorative and non-interactive elements

Session Isolation & Resource Management

Multi-Session Browser Isolation

Each MCP client session now gets its own isolated browser environment:

• Session-based isolation - Browser sessions are keyed by client session ID
• Automatic cleanup - Sessions are released when clients disconnect
• No cross-contamination - Multiple clients can work independently

📊 Observability & Client Tracking

Client Identification & Logging

Know exactly which AI clients are using your tools:

• Client name detection - Automatically identifies connecting clients (Claude, GPT, etc.)
• Request tracking - All tool calls logged with client information
• Usage attribution - Understand which clients generate which requests

Logging Example:

[search_engine] executing (client=claude-desktop) {"query": "AI news"}

Error Handling Improvements

• Polling optimization - Dataset polling now stops immediately on HTTP 400 errors instead of retrying
• Better error messages - More descriptive error context for debugging
• Graceful degradation - Improved handling of browser disconnections

Fixed Values Support for Datasets

Dataset tools now support fixed parameter values:

• Reduces token count in tool definitions
• Simplifies tool usage for AI models
• Pre-configures common parameters

---

🔄 Breaking Changes

Tool Renaming

Some tools have been renamed. If you have hardcoded tool names, please update:

• Check tool descriptions for any name changes
• Update any automation scripts accordingly

---

🐛 Bug Fixes

• Dataset polling - Fixed infinite retry loop on permanent failures (HTTP 400 now stops polling immediately)
• Browser session cleanup - Fixed memory leaks from unclosed sessions
• Error propagation - Improved error context in deeply nested calls
• Connection handling - Better recovery from browser disconnections

---

🔮 Looking Ahead

This release lays the groundwork for exciting future enhancements:

• Advanced browser automation capabilities
• Enhanced analytics and usage insights
• More dataset integrations
• Performance optimizations

---

📚 Resources

• Documentation: https://docs.brightdata.com/mcp-server/overview
• GitHub Repository: https://github.com/brightdata/brightdata-mcp
• Support: Contact us at support@brightdata.com

---

🐛 Report Issues

Found a bug or have a feature request? We'd love to hear from you!

• Open an issue on GitHub
• Contact support at support@brightdata.com
• Join our community channels for discussions

---

Happy Automating! 🚀

The Bright Data Team

Full Changelog: v2.4.2...v2.6.0