Bright Data
Also known as: Bright Data Web MCP, Luminati Networks, Web Unlocker, Scraping Browser
Web data infrastructure for agents, spanning a 400 million IP residential network, Web Unlocker, Scraping Browser and a Web MCP server with a free tier that lets Claude, Cursor and Windsurf navigate and act on the live web unblocked.
Bright Data is the web data infrastructure most agents end up standing on, whether or not their builders name it. Founded as Luminati Networks and rebranded in 2021, it runs a residential proxy pool exceeding 400 million IPs with city, ASN and carrier targeting, and has expanded into a full stack: Web Unlocker, Scraping Browser, SERP API, Scraper APIs and Studio, ready to use datasets, real time feeds, a CLI and managed data acquisition. More than 20,000 customers including Fortune 500 companies use it, and G2 ranks it the leading proxy network across 240 plus verified reviews.
For agents specifically, its Web MCP server lets language models and autonomous agents access, discover and extract web data in real time. MCP clients including Claude Desktop, Cursor and Windsurf can search the web, navigate sites, take actions and retrieve data without being blocked, and there is a free tier covering 5,000 requests a month, which is an unusually open on ramp for infrastructure of this class. It also supplies multimodal training data, including real world video for vision language action models, robotics and world model work.
Its strongest and least replicable asset is legal. Bright Data won two landmark cases, Meta versus Bright Data and X Corp versus Bright Data, affirming the legality of collecting public web data, and it gates residential access behind a mandatory know your customer process. Compliance follows: SOC 2 Type II with a publicly downloadable SOC 3, ISO 27001:2022 alongside 27017, 27018 and 27701, CSA STAR Level 1, GDPR and CCPA with data processing agreements, and a published ethics report covering its opt in peer network.
The honest counterweights are cost and billing clarity. Pay as you go residential runs around $5 per gigabyte, which reviewers put at two to three times budget providers, and the billing rules differ by product in ways the company does not surface at signup: Web Unlocker bills on success by default but charges every request once custom headers or cookies are enabled, Scraping Browser charges per gigabyte including bandwidth from failed attempts, and SERP API does not bill failures. Onboarding friction is real too, with government ID upload required and freelancers reportedly asked to complete an English language video interview.
Vendor details
Canonical URL
https://brightdata.com
Category
Agent infrastructure
Company status
unclear
Use cases & customers
Target customers
Deployment options
In practice
Your agent needs the live web and keeps getting blocked. The Web MCP server gives Claude, Cursor or Windsurf unblocked search, navigation and extraction, with 5,000 requests a month free to start.
Your legal team asks whether collecting this data is defensible. Bright Data won two landmark cases affirming the legality of public web data collection and publishes an ethics report and KYC gated sourcing.
Procurement needs certifications rather than assurances. SOC 2 Type II with a downloadable SOC 3, ISO 27001:2022, 27017, 27018, 27701 and CSA STAR Level 1 are all published.
Sources & related URLs
Agentic Index coverage score
9.5 / 14 capabilities · 68%
| Integrations & Tool CallingWeb Access APIs, Scraper APIs, Scraper Studio, ready to use datasets, real time data feeds, MCP, CLI and managed data acquisition, plus AWS Marketplace procurement so purchases draw down existing AWS commitments. 2026-08-05 | Full |
|---|---|
| Workflow OrchestrationSupports every stage of collection from access and extraction through structuring, automation and delivery, letting teams build cloud data pipelines without maintaining infrastructure, but this is data pipeline orchestration rather than agent orchestration; the control loop belongs to whatever calls it. 2026-08-05 | Partial |
| Knowledge Grounding & RAGSupplies the live web as grounding material for agents and ships curated licensed datasets through AI Data Launchpad, but it provides no retrieval index, vector store or citation layer of its own; it is the source that grounds, not the grounding layer. 2026-08-05 | Partial |
| Human Oversight & GuardrailsPer zone spend limits set in bytes or dollars and evaluated every fifteen minutes, plus sub user permissions, give real operational guardrails on autonomous consumption, and KYC gates who can use the residential network at all, but there is no approval workflow over agent actions. 2026-08-05 | Partial |
| Security, Identity & GovernanceThe strongest compliance stack in this lane: SOC 2 Type II with a publicly downloadable SOC 3, ISO 27001:2022 alongside ISO 27017, 27018 and 27701, CSA STAR Level 1, GDPR and CCPA with data processing agreements, third party audits, a published ethics report and mandatory KYC verification gating residential access. 2026-08-05 | Full |
| Observability & AuditabilityA control panel centralising zone creation, sub user permissions, real time usage analytics and per zone spend limits evaluated every fifteen minutes, plus a public network status page and a 99.99 percent uptime SLA on enterprise tiers. Cost and consumption visibility is first class, which matters when agents drive the spend. 2026-08-05 | Full |
| Memory & State PersistenceISP proxies support sticky account sessions that persist across requests and datasets persist as durable artifacts, but no agent scoped memory or cross run state model is provided. 2026-08-05 | Partial |
| Deployment & Data ResidencyThe MCP server is documented as a local server the customer runs, and cloud delivery options plus global infrastructure are offered, but the collection network itself is inherently vendor operated and no regional data residency commitment is documented. 2026-08-05 | Partial |
| Prebuilt Agents, Templates & PacksReady to use datasets and a dataset marketplace, prebuilt scraper APIs per target, Scraper Studio, real time feeds, and the AI Data Launchpad programme of curated licensed datasets for AI and machine learning use cases. 2026-08-05 | Full |
| Triggers & Channel CoverageReachable through APIs, CLI and MCP with real time data feeds pushing updates, which is broad protocol coverage, but the platform is invoked rather than triggering work itself and no scheduling or event surface is documented. 2026-08-05 | Partial |
| Model Flexibility & RoutingNo models are involved. Bright Data supplies web data to whatever model the customer runs, which is the correct architectural boundary for this component but scores nothing on this axis. 2026-08-05 | Unable to verify |
| APIs, SDKs & MCP ExtensibilityA documented Web MCP server with its own free tier of 5,000 requests per month, alongside Web Access APIs, Scraper APIs, a CLI and AWS Marketplace distribution. The MCP surface is a first class product with its own pricing page rather than an afterthought. 2026-08-05 | Full |
| Testing, Debugging & OptimizationIndependent benchmarking tracks ban rate and pool freshness, with reported sub one percent bans against Cloudflare fronted retailers and 94 percent of IPs unseen in the prior week, and a Scraper APIs playground supports prototyping, but no evaluation harness for agent behaviour is provided. 2026-08-05 | Partial |
| Browser & Computer UseThe Scraping Browser provides real browser automation, and the MCP server explicitly lets agents search the web, NAVIGATE WEBSITES AND TAKE ACTIONS, with Web Unlocker handling JavaScript rendering, CAPTCHA solving and geo restriction bypass so those actions complete against defended targets. 2026-08-05 | Full |
The Agentic Index coverage score grades every vendor Full, Partial or Unable to verify against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Pricing
Web MCP free tier at 5,000 requests per month; pay as you go residential around $5.04 per GB; volume commitments negotiated below sticker
gigabytes transferred, requests, or successful results depending on product
What is public
The MCP free tier allowance and per GB residential rates are published, and compliance documentation including a downloadable SOC 3 is public. Enterprise and volume rates are negotiated.
Billing mechanics
Usage based across a product family with different units. The Web MCP server carries its own free tier of 5,000 requests a month. Residential proxy access is roughly $5 per gigabyte pay as you go, with volume commitments negotiated well below sticker above about a terabyte a month. Enterprise tiers add unlimited seats, a free Proxy Manager licence, dedicated account managers and a 99.99 percent uptime SLA, and purchases can be routed through AWS Marketplace to draw down existing AWS commitments.
Cost watchouts
Billing rules differ by product and are not surfaced at signup: Web Unlocker bills on success by default but charges EVERY request once custom headers, cookies or expect elements are enabled; Scraping Browser charges per GB including bandwidth from failed attempts; SERP API does not bill failures. Calendar month commitments and non rolling balances also generate unexpected top up charges.
Variable cost rationale
Cost tracks bandwidth and request volume, both of which an autonomous agent controls rather than the buyer, and two of the four billing modes charge for failed attempts, so a poorly behaved agent hitting defended targets can consume budget without producing results.
Additional watchouts
Two distinct frictions beyond price. First, the per product billing differences above, which the vendor does not make obvious at signup and which reviewers cite as the most common complaint. Second, onboarding: KYC requires government ID upload, only registered companies can complete full verification, and freelancers are reportedly asked to complete an English language video interview, with official 48 hour approval running from three days to three weeks in practice.
Overage / add-ons
Consumption metered per product: per GB for residential and Scraping Browser, per successful request for Web Unlocker in default mode, per request for SERP API
Sales call required
Mixed (some tiers require a call)
Free / trial
Web MCP server free tier covering 5,000 requests a month for scraping search results and unlocking public pages; a separate three day product trial with limited bandwidth and restricted features
Lowest paid plan
Pay as you go residential
Commercial notes
The free MCP tier is strategically interesting: it makes Bright Data the default web access layer inside Claude Desktop, Cursor and Windsurf at no cost, which is distribution rather than pricing. Against that, reviewers consistently describe the platform as two to three times budget providers and in some cases three to ten times, with one summarising it as bringing a firehose to water a houseplant for small users. The premium buys compliance and success rates rather than raw access.
Key ambiguities
Rates vary by product and volume tier and most teams above a terabyte a month negotiate below the published sticker, so the public figure is a ceiling. The relationship between the free MCP tier and paid product consumption once limits are exceeded is not fully documented.
Missing data
Full per product rate card, enterprise commitment thresholds, and how MCP free tier usage converts to billable consumption at scale.
Related vendors
- Acrab — Singapore compute infrastructure company building a full stack…
- AgentOps — Agent observability and reliability platform with broad model and…
- Agno — High-performance agent runtime and framework (formerly Phidata) with…
- AIsa — Unified resource and payment gateway for AI agents that lets them…
- AlphaBitCore — AI control plane that governs how models, agents, tools, and…
- Anchor Browser — Cloud hosted browser infrastructure that lets AI agents operate real…
Alternatives to Bright Data
The closest documented capability profiles to Bright Data among agent infrastructure platforms tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Apify9.0 / 14A lighter documented profile than Bright Data
- Browserless9.5 / 14Fuller documented coverage on Memory & State Persistence and Deployment & Data Residency
- Hyperbrowser7.0 / 14A lighter documented profile than Bright Data
- Steel8.0 / 14Fuller documented coverage on Deployment & Data Residency
- BotGauge8.5 / 14Fuller documented coverage on Workflow Orchestration and Human Oversight & Guardrails
- Kong9.5 / 14Adds documented Model Flexibility & Routing
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded