Building Web Search Agent with Google-ADK

Set up a web search agent with Google ADK and LangDB. Guide covers creating virtual MCP servers, integrating search tools, model routing, and full observability.

This guide walks you through setting up a powerful WebSearch agent that leverages Google ADK for orchestration and LangDB for LLM access, tracing, and flexible routing.

Checkout: https://app.langdb.ai/sharing/threads/28cf3e91-956d-4b32-a4dd-ed00bfdf3a1c

Code

• LangDB Samples: https://github.com/langdb/langdb-samples/tree/main/examples/google-adk/web-search-agent

Overview

• Google ADK: Manages agent workflows and tool execution.

• LangDB: Provides access to 350+ models (OpenAI, Anthropic, Gemini, etc.), with full tracing, tool integration, and cost controls.

• MCP/Tools: Use LangDB Virtual MCP system to integrate external tools like Tavily Search into your agent’s workflow.

You’ll learn how to:

• Configure a Google ADK agent to use LangDB as its LLM provider

• Set up and use MCP tool calls (e.g., for real-time web search)

• Route and trace all agent activity using LangDB

Agents Involved

Critic Agent (critic_agent)

• Role: Web search and analysis

• Step: First in a 2-step pipeline

• Functionality:

  • Receives the user query

  • Performs a web search using MCP tools (e.g., Tavily)

  • Annotates its response with grounding references (links, citations)

Reviser Agent (reviser_agent)

• Role: Synthesis and refinement

• Step: Second in the pipeline

• Functionality:

  • Takes the output from Critic Agent

  • Refines, synthesizes, and clarifies the response

Architecture Overview

• Two-Step Pipeline:

  1. Critic Agent analyzes and gathers evidence via web search

  2. Reviser Agent polishes the answer for delivery

• Shared Context: Both agents use shared thread/run IDs to maintain continuity

• Extensible: This design can add more agents or steps as needed

• Tech Stack: Built on LlmAgent from google.adk.agents, using LangDB LLM interface and MCP integration

User Query
   │
   ▼
[Critic Agent]  --(web search, analysis, references)-->
   │
   ▼
[Reviser Agent] --(synthesis, refinement)-->
   │
   ▼
Final Answer to User

Enabling Web Search Agents with Virtual MCP Server

To enable real-time web search for your agent, you must use a Virtual MCP Server in LangDB. This ensures all web search tools (such as Tavily) are available and can be integrated seamlessly into your agent's workflow.

Steps to Create a Virtual MCP Server

• Click + New Virtual MCP Server:

  • Name: web-search-mcp

  • Underlying MCP: Tavily Search MCP

    • Requires API Key: Make sure Tavily API Key is configured in your environment to authenticate this operation.

• After Creating a Virtual MCP Server

  • You get the MCP Url from the Code Snippets.

  • For example:

    

We will be using this MCP Server URL in the Critic Agent:

server_url = "https://api.us-east-1.langdb.ai/mcp/tavily_4ykdv5fj"
critic_agent = LlmAgent(
    model=LangDBLlm(
        model="openai/gpt-4.1",
        api_key=os.getenv("LANGDB_API_KEY"),
        project_id = os.getenv("LANGDB_PROJECT_ID"),
        extra_headers={
            "x-thread-id": SHARED_THREAD_ID,
            "x-run-id": SHARED_RUN_ID
        },
        mcp_servers = [{
            "server_url": server_url,
            "type": "sse",
            "name": "Tavily"
        }]
    ),
    name="critic_agent",
    instruction=prompt.CRITIC_PROMPT,
    after_model_callback=_render_reference,
)

Running the Agent

Pre-requesites:

Install Google ADK and LangDB :

pip install google-adk langdb-adk

What is langdb-adk?

langdb-adk is an extension package for Google ADK. It provides additional utilities and integrations that make it easier to use LangDB as your LLM backend within Google ADK workflows.

Executing the Agent:

adk run web-search
<!-- or -->
adk web

Output

Here is a conversation with the Web-Search agent:

AI Conversation | LangDBLangDB

References

• Google ADK Documentation

• LangDB Virtual MCP Servers

• LangDB Virtual Models

• Tavily Search MCP