Bright Data

Agents get web data as tools: the MCP server's 60-plus tools cover search, page unlocking and structured extractors for sites such as Amazon, LinkedIn and Instagram, the Web Scraper API offers 1,300-plus pre-built scrapers, and Scraper Studio builds custom scrapers from a prompt. These read public web data; they take no actions in the customer's own systems and hold no per-user credentials to scope.

Sourcedocs.brightdata.com/products/mcp-server/toolsread 2026-09-22

What runs here are collection jobs: a scraper is triggered, runs on Bright Data's infrastructure and delivers results. No way to sequence, branch or route multiple steps into a workflow for the customer is documented; the vendor's tutorials chain SERP and scraper calls in the customer's own code or GitHub Actions.

Sourcedocs.brightdata.com/llms.txtread 2026-09-22

Public web data, ready-made datasets and a people and company search index that Bright Data maintains are what it supplies; none of these is a retrieval structure over the customer's own documents or data. No ingestion, retrieval or RAG layer over company data is documented, and public web content is not the customer's corpus.

Sourcedocs.brightdata.com/llms.txtread 2026-09-22

A person approves AI changes before they commit: when Scraper Studio's AI self-healing rewrites a broken scraper, the heal stops by default at an approval gate with status awaiting_approval and a preview of the result, and the fix commits only when the user approves it (or is rejected). Per-zone spend limits and zone permissions on API keys add constraints.

Sourcedocs.brightdata.com/products/cli/commandsread 2026-09-22

Bright Data publishes ISO/IEC 27001:2022, 27017 and 27018 certificates (numbers and validity dates given) scoped to its web data platform including the MCP server and Browser API, a SOC 2 Type II report under NDA and a publicly downloadable SOC 3 from a Deloitte network firm, a CSA STAR listing, and GDPR and CCPA programs. On the access side, API keys are bound to zones with their own permissions and admin rights, and residential proxies require a KYC-verified business account.

Sourcedocs.brightdata.com/general/security/security-overviewread 2026-09-22

Usage is recorded: the control panel's Cost Explorer and zone statistics break usage down by product, zone and hour, and the self-hosted Proxy Manager can export request logs to a webhook, Datadog, Logz.io or S3. A record of what an agent asked for and received through the MCP server or the APIs, step by step, is not documented.

Sourcedocs.brightdata.com/products/proxy-manager/faqsread 2026-09-22

Bright Data keeps collection results as job snapshots for download and holds sessions for browsing, but no memory layer that keeps an agent's context across runs is documented.

Sourcedocs.brightdata.com/llms.txtread 2026-09-22

Collection runs on Bright Data's network, but results can land in the customer's own infrastructure (S3, Google Cloud Storage, Azure, Snowflake, SFTP or a webhook), and the MCP server and the open-source Proxy Manager can run locally. No region choice for where Bright Data processes and stores data, and no self-hosted collection platform, is documented.

Sourcedocs.brightdata.com/products/scrapers/scrapers-library/data-deliveryread 2026-09-22

Tools come packaged by use case: the MCP server offers 11 pre-configured groups (e-commerce, social media, browser automation, business intelligence, finance, research and others), and Bright Data publishes Agent Skills for coding agents. The 1,300-plus pre-built scrapers are integrations, and the 350-plus datasets are data products rather than workflows. These are vendor made bundles for defined jobs; no catalog of ready made workflows or agents is published.

Sourcedocs.brightdata.com/products/mcp-server/toolsread 2026-09-22

Collection can start and report without a person: Scraper Studio scrapers run through initiation endpoints that include a schedule, and scraper results are pushed to the customer's webhook for event-driven pipelines, or to S3, Google Cloud Storage or Snowflake, when a job completes. The schedule is documented only as one of the initiation modes.

Sourcedocs.brightdata.com/products/scrapers/scrapers-library/data-deliveryread 2026-09-22

Any model the customer's agent runs can take Bright Data's web data: the MCP server works with any MCP client, and the docs cover integrations with coding agents and agent frameworks, each using the customer's own model. Bright Data names no model of its own for agents, so model choice is the customer's.

Sourcedocs.brightdata.com/products/mcp-server/overviewread 2026-09-22

Documented REST APIs cover every product (Web Scraper, Scraper Studio, Web Unlocker, SERP, account management and dataset marketplace), alongside Python and JavaScript SDKs, a CLI, Agent Skills, a remote and local MCP server with OAuth, and a protocol that lets an agent register an account without a signup form.

Sourcedocs.brightdata.com/llms.txtread 2026-09-22

Bright Data documents error codes, previews of self-healed scrapers and fill-rate statistics for its datasets, which check its own scrapers and data; no way to test or score the customer's agent is documented.

Sourcedocs.brightdata.com/llms.txtread 2026-09-22

The Browser API runs the customer's Playwright, Puppeteer or Selenium code in fully managed cloud browsers built for user-like interactions, and Bright Data's MCP server gives an agent browser tools to navigate, click, type, scroll, wait for elements and take screenshots.

Sourcedocs.brightdata.com/products/scraping-browser/introductionread 2026-09-22