Providing Context to the Agent

[Syth03]

I have made an Agent with the help of the MCP Server. But now, I want that my agent should have some additional context and for that, I want to provide it a knowledge base.. Just as how we have in CrewAI and Agno. How am I supposed to do that?

[dexhunter]

Add a retrieval tool alongside your MCP toolset. ADK does not require the knowledge base to live in the MCP server: it can be a separate index that your agent searches when it needs background information.

The integration looks like this:

from google.adk.agents import Agent

def search_knowledge(query: str) -> list[dict]:
    """Find relevant passages in the project knowledge base."""
    return retriever.search(query)

root_agent = Agent(
    name="assistant",
    model="gemini-2.5-flash",
    instruction=(
        "Use search_knowledge for questions about the project. "
        "Base your answer on the returned passages and cite their sources."
    ),
    tools=[mcp_toolset, search_knowledge],
)

Here retriever is your own retrieval implementation and mcp_toolset is the MCP toolset you already configured. Have the retriever return JSON-serializable records such as {"text": "relevant passage", "source": "document URL or title"}.

Prepare the documents outside the agent loop: extract text, split it into passages, and index those passages for keyword or vector search. The tool returns only the relevant passages for the current question. ADK's FunctionTool handles exposing a Python function as a callable tool.

For a small, fixed amount of context, you can instead include it in instruction. Conversation memory and a document knowledge base serve different purposes; adding a memory service alone does not ingest your project documents.