Gemini Interactions - Deep Research with MCP servers

Give the Deep Research agent access to external tools via remote MCP servers. Pass server configs through mcp_servers; type: "mcp_server" is added automatically. Only url is strictly required; name, headers, and allowed_tools are optional.

deep_research_mcp.py

"""
Gemini Interactions - Deep Research with MCP servers
=====================================================

Give the Deep Research agent access to external tools via remote MCP servers.
Pass server configs through `mcp_servers`; `type: "mcp_server"` is added
automatically. Only `url` is strictly required; `name`, `headers`, and
`allowed_tools` are optional.

Custom Function Calling tools are NOT supported by Deep Research, but remote
MCP servers are.
"""

from agno.agent import Agent
from agno.models.google import GeminiInteractions

agent = Agent(
    model=GeminiInteractions(
        agent="deep-research-preview-04-2026",
        thinking_summaries="auto",
        mcp_servers=[
            {
                "name": "Deployment Tracker",
                "url": "https://mcp.example.com/mcp",
                "headers": {"Authorization": "Bearer my-token"},
                # "allowed_tools": ["get_status"],  # optionally restrict
            }
        ],
    ),
    markdown=True,
)

if __name__ == "__main__":
    agent.print_response(
        "Check the status of my last server deployment and summarize any issues."
    )

Run the Example

Set up your virtual environment

uv venv --python 3.12
source .venv/bin/activate

Install dependencies

uv pip install -U agno google-genai

Export your Google API key

export GOOGLE_API_KEY="your_google_api_key_here"

Configure the remote MCP server

Replace https://mcp.example.com/mcp and Bearer my-token with your reachable server URL and issued credential. Set allowed_tools to the deployment-reading tools you intend to expose. Google connects to the remote server; this path does not start a local MCP process. Read secrets from your application's secret configuration instead of committing them.

Run the example

Save the code above as deep_research_mcp.py, then run:

python deep_research_mcp.py

Full source: cookbook/90_models/google/gemini_interactions/deep_research_mcp.py