Gemini Provider

Google Gemini models with native function calling and full multimodal support (text, image, audio, video, file).

Install

bashpip install "mtpx[gemini]"

Or install the SDK directly:

bashpip install google-genai

API Key Setup

Option 1: .env file (recommended)

1. Install dotenv support:

bash   pip install python-dotenv

Or with the MTP extra:

bash   pip install "mtpx[dotenv]"

1. Create a .env file in your project root:

text   GEMINI_API_KEY=AIzaSyour_key_here

1. Load it in your code before creating the provider:

python   from mtp import Agent

   Agent.load_dotenv_if_available()  # reads .env file

Get a free API key at aistudio.google.com.

Option 2: System environment variable

bash# Linux/macOS
export GEMINI_API_KEY="AIza..."

# Windows PowerShell
$env:GEMINI_API_KEY="AIza..."

Quick Start

pythonfrom mtp import Agent
from mtp.providers import Gemini

Agent.load_dotenv_if_available()  # loads GEMINI_API_KEY from .env

provider = Gemini(model="gemini-2.0-flash")
tools = Agent.ToolRegistry()
agent = Agent(provider=provider, tools=tools)

reply = agent.run_loop("What is 25 * 4 + 10?")
print(reply)

Parameters

Capabilities

Recommended Models

• gemini-2.0-flash — Fast, good tool calling (default)
• gemini-2.5-pro — Most capable, best reasoning
• gemini-2.0-flash-lite — Cheapest option

Multimodal Support

Gemini supports all modalities natively:

pythonfrom mtp.media import Image, Audio, Video, File

# With image
reply = agent.run_loop({
    "content": "Describe this image",
    "images": [Image(filepath="photo.jpg")],
})

# With audio
reply = agent.run_loop({
    "content": "Transcribe this audio",
    "audios": [Audio(filepath="speech.mp3")],
})

# With video
reply = agent.run_loop({
    "content": "What happens in this video?",
    "videos": [Video(filepath="clip.mp4")],
})

Full Example

pythonfrom mtp import Agent
from mtp.providers import Gemini

Agent.load_dotenv_if_available()

provider = Gemini(
    model="gemini-2.0-flash",
    temperature=0.0,
)

tools = Agent.ToolRegistry()
agent = Agent(provider=provider, tools=tools, debug_mode=True)

reply = agent.run_loop(
    "Calculate (25 * 4) + 10 and list files in the current directory.",
    max_rounds=4,
)
print(reply)

Notes

• Gemini uses function_declarations in tools (not the OpenAI function wrapper format). MTP handles the translation automatically.
• Tool schemas are sanitized to remove unsupported JSON Schema keys before sending to Gemini.
• Parallel tool calls are not supported by Gemini's API as of now.

Source

src/mtp/providers/gemini_provider.py