22 SepBuilding & Governing Banking Agents: Arcade & Mastra

Web Research Agent with Bright Data

Ask a question or provide a public URL. The agent searches the web, reads relevant pages, and returns an answer with source links. The research-brief workflow returns a summary, key points, and sources as structured JSON. Built with Mastra and Bright Data’s hosted MCP server.

Why we built this

Web research requires both finding relevant pages and retrieving their content. Some sites block ordinary HTTP requests or return incomplete pages. This template uses Bright Data for search and page retrieval, with a conversational agent for questions and a workflow for structured research briefs. Retrieval can still fail, and answers depend on the sources available.

Demo

Prerequisites

  • OpenAI API key: set OPENAI_API_KEY for the default model, openai/gpt-5.6-luna. You can change the model in the agent configuration; another provider may require different credentials.
  • Bright Data API token: set BRIGHT_DATA_API_TOKEN for web search and page scraping.

Quickstart 🚀

  1. Clone the template
    • Run npx create-mastra@latest --template https://github.com/mastra-ai/template-bright-data to scaffold the project locally.
    • Enter the generated project directory and run npm install if dependencies were not installed during setup.
  2. Add your API keys
    • Run cp .env.example .env and fill in the values described under Prerequisites.
  3. Start the dev server
    • Run npm run dev and open localhost:4111, or the Studio URL printed in your terminal.
    • Select Web Agent and try the product-page prompt below. The agent should return available product details with a source link.

Try it out

  • Ask Web Agent: “Read https://www.amazon.com/dp/B0BDHWDR12 and report the product name, current price, and star rating. Include the source URL and identify any fields you cannot retrieve.” Look for details supported by the page and an explanation of anything missing. Product availability and page content may vary.
  • Ask Web Agent: “Compare SQLite and PostgreSQL for a small web application using their official documentation. Read the relevant pages and include source links.” Look for a comparison based on retrieved pages, with links you can inspect.
  • Run research-brief with {"topic":"What are the differences between SQLite and PostgreSQL for a small web application? Use their official documentation."}. The workflow first searches and reads sources, then writes a brief from those findings with no further web calls. The JSON result contains topic, summary, keyPoints, and source titles and URLs.

Customization

  • Open the project in your coding agent and ask: “Adapt the research brief to compare a list of product URLs, returning available prices, specifications, and source links. Explore the code and propose a plan before making changes.” The output schema and research prompts are in the research workflow.
  • Connect the agent or workflow to an application using the Mastra Client SDK.

Platform-specific tools

Set BRIGHT_DATA_MCP_GROUPS in .env and restart the dev server to enable a platform group:

BRIGHT_DATA_MCP_GROUPS=ecommerce
GroupTools for
ecommerceAmazon, Walmart, eBay, Best Buy, Etsy, Home Depot, Zara, Google Shopping
socialLinkedIn, Instagram, TikTok, YouTube, X, Reddit, Facebook
businessCrunchbase, ZoomInfo, Google Maps reviews, Zillow, Booking.com
browserRemote browser navigation, clicks, typing, scrolling, and screenshots
geoChatGPT, Grok, and Perplexity answers as structured data

The hosted server has these configuration constraints:

  • One group applies per connection. A comma-separated list uses only the first entry.
  • Enabling a group removes the search_engine_batch and scrape_batch tools.
  • An unrecognized group name is ignored, leaving the default tools available.

Leave the variable empty unless you need a particular group.

For an alternative to the hosted MCP connection, @mastra/brightdata provides search and page-fetching tools through the Bright Data SDK. It requires Bright Data zones configured on your account and does not include the MCP batch tools or platform groups described above.

Tool connection errors

Studio can start without a Bright Data token, but research requests require working web tools. If tool discovery fails, the request reports an error and the next request tries discovery again. Restart the server after changing your API keys.

About Mastra templates

This partnership template was contributed by Bright Data to show how Mastra uses Bright Data’s search and page-retrieval tools for web research.

Want to contribute?