Give Your AI Agents
Live Web Access
An agent that can't read the web is guessing. DataBlue ships a native MCP server plus a clean REST API, so Claude, Cursor, and your own agents can search, scrape, map, and crawl the web as tool calls, with no glue code or proxy plumbing.
Web Tools Agents Actually Use.
Nine Tools, One Server
Point any MCP-compatible client, Claude Desktop, Cursor, or your own runtime, at DataBlue and your agent instantly gains a full set of web tools, covered below. No custom integration layer to build or maintain.
Structured, Token-Lean Output
Agents pay for every token they read. DataBlue returns clean markdown and typed JSON instead of raw HTML, so context windows stay focused on signal, not nav bars, scripts, and boilerplate.
Search + Scrape in One Tool Belt
Discovery and extraction live behind the same API key. An agent can search the web, pick a result, and read the full page in a single reasoning loop, exactly the pattern tool-using LLMs are built for.
Reliable Under Autonomy
Agents retry, branch, and run unattended. Plan-based concurrency, JavaScript rendering, and non-expiring top-up credits make long-running agent workflows easier to budget.
Prebuilt Agent Workflows
Six MCP prompts ship ready to use, covering company research, competitor SERP reports, and lead finding, among others. An agent runs a whole workflow from a single prompt call instead of you writing the reasoning chain by hand.
The Data API Catalog, as Tools
Beyond search and scrape, the server exposes DataBlue's Data API catalog through two tools, one to list what's available and one to run it. An agent reaches commerce, quick-commerce, or Google data sources without a separate integration for each.
Add DataBlue as an MCP Server.
Drop this into a JSON-based MCP client config, like Claude Desktop, Cursor, or Cline, and your agent gains web tools on the next restart. Codex uses an equivalent TOML block, covered in the docs.
{
"mcpServers": {
"datablue": {
"command": "npx",
"args": ["--yes", "--prefer-online", "@datablue/mcp@latest"],
"env": {
"DATABLUE_API_KEY": "<your-key>"
}
}
}
}
// Your agent can now call:
// datablue_search_google -> live Google results
// datablue_scrape_url -> page as markdown / JSON
// datablue_batch_scrape -> up to 20 URLs, one call
// datablue_map_site -> discover a site's URLs
// datablue_start_crawl -> whole-site extraction
// datablue_get_job_status -> poll a crawl or search job
// datablue_extract -> structured fields, one prompt
// datablue_list_data_apis -> browse the Data API catalog
// datablue_run_data_api -> run a catalog API by idFrom Config to Web-Aware Agent.
Connect the MCP Server
Add the DataBlue MCP server to Claude, Cursor, or your own agent runtime with a single config block and your API key.
Agent Calls the Tools
Your agent decides when to search and scrape, calling the tools mid-reasoning just like any other function. There's no orchestration code from you.
Grounded Responses
Results come back as clean, structured data the model can read directly, so the agent answers from the live web instead of guessing.
AI Agent Questions.
What is MCP and why does it matter for agents?
MCP (Model Context Protocol) is an open standard for connecting tools to AI agents. Because DataBlue ships a native MCP server, any MCP-compatible client, like Claude Desktop or Cursor, gets web search and scraping as built-in tools, without you writing a custom integration.
Which agents and clients work with DataBlue?
Any client that supports local stdio MCP can connect, including Claude Desktop, Claude Code, Cursor, Codex, Cline, and Windsurf-style clients. Claude Desktop, Cursor, and Cline read the same mcpServers JSON shown above; Codex uses an equivalent TOML config instead. If you'd rather not use MCP, the same capabilities are available through the standard REST API.
What tools does the MCP server actually expose?
Nine of them. datablue_search_google and datablue_scrape_url cover the basics. datablue_batch_scrape and datablue_extract handle multiple URLs at once, up to 20 per call. datablue_map_site and datablue_start_crawl (paired with datablue_get_job_status) handle site-wide discovery and crawling. datablue_list_data_apis and datablue_run_data_api open up DataBlue's full Data API catalog as a callable tool pair.
Does DataBlue include prebuilt agent workflows?
Yes, six MCP prompts ship with the server: research_company, competitor_serp_report, extract_products_from_url, find_leads_from_google, scrape_site_to_markdown, and monitor_keyword. Each one chains the underlying tools into a working prompt, so an agent runs the whole workflow from a single call instead of you scripting the steps.
Why not just let the agent fetch URLs itself?
Raw fetching breaks on JavaScript-rendered pages, returns noisy HTML that wastes tokens, and gets blocked without proxy infrastructure. DataBlue renders the page, strips the noise, and returns clean markdown or JSON, so the agent reads signal, not boilerplate.
How does this handle pages that block bots?
DataBlue manages rendering and anti-bot handling for you behind the API, so your agent can read pages that would otherwise return empty or blocked responses, no proxy pool or browser farm to operate.
Will autonomous runs blow through my quota?
Pricing is credit-based and top-up credits do not expire, so unattended runs don't waste a resetting monthly allowance. You can start with 1,000 one-time signup credits and scale up with predictable endpoint weights.
Ship an Agent That Reads the Web.
Start with 1,000 one-time signup credits, no card required. Connect the MCP server and give your agent live web tools in minutes.