AgentQL¶

The AgentQL MCP Server connects your ADK agent to AgentQL. AgentQL is a semantic extraction engine that queries web elements based on their meaning rather than their CSS or XPath selectors. This functionality allows agents to retrieve specific data points from web pages, PDFs, and authenticated sessions using natural language definitions.

Use cases¶

Resilient Web Extraction: Extract data from dynamic websites using natural language descriptions. This feature allows your agent to reliably gather information from sites that frequently update their layout or CSS without breaking.
Data Normalization: Convert unstructured web pages into clean, predictable JSON formats. This capability enables your agent to instantly normalize data from different sources (like multiple job boards or shopping sites) into a single schema.

Prerequisites¶

Create an API Key in AgentQL. Refer to the documentation for more information.

Use with agent¶

Local MCP Server

from google.adk.agents import Agent
from google.adk.tools.mcp_tool.mcp_session_manager import StdioConnectionParams
from google.adk.tools.mcp_tool.mcp_toolset import MCPToolset
from mcp import StdioServerParameters

AGENTQL_API_KEY = "YOUR_AGENTQL_API_KEY"

root_agent = Agent(
    model="gemini-2.5-pro",
    name="agentql_agent",
    instruction="Help users get information from AgentQL",
    tools=[
        MCPToolset(
            connection_params=StdioConnectionParams(
                server_params = StdioServerParameters(
                    command="npx",
                    args=[
                        "-y",
                        "agentql-mcp",
                    ],
                    env={
                        "AGENTQL_API_KEY": AGENTQL_API_KEY,
                    }
                ),
                timeout=300,
            ),
        )
    ],
)

Available tools¶

Tool	Description
`extract-web-data`	Extract structured data from a given 'url', using 'prompt' as a description of actual data and its fields to extract

Best practices¶

To ensure accurate extraction, follow these guidelines when prompting the agent:

Describe the data, not the element: Avoid visual descriptions (e.g., "the blue button"). Instead, describe the data entity (e.g., "the submit button" or "the product price").
Define the hierarchy: If extracting a list, explicitly instruct the agent to look for a collection of items and define the fields required for each item.
Filter semantically: You can instruct the tool to ignore specific data types (e.g., "exclude ads and navigation links") within the prompt itself.