Cohere Provider

Cohere provides Command models with native RAG grounding and multi-step agentic tool use built into the model.

Install

bashpip install "mtpx[cohere]"

Or install the SDK directly:

bashpip install cohere

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   COHERE_API_KEY=your_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 trial key at dashboard.cohere.com (no credit card required).

Option 2: System environment variable

bash# Linux/macOS
export COHERE_API_KEY="..."

# Windows PowerShell
$env:COHERE_API_KEY="..."

Quick Start

pythonfrom mtp import Agent
from mtp.providers import Cohere

Agent.load_dotenv_if_available()  # loads COHERE_API_KEY from .env

provider = Cohere(model="command-a-03-2025")
tools = Agent.ToolRegistry()
agent = Agent(provider=provider, tools=tools)

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

Parameters

Capabilities

Recommended Models

• command-a-03-2025 — Most capable, best tool use (default)
• command-r-plus-08-2024 — Strong RAG + multi-step tool use
• command-r-08-2024 — Fast, cheaper, solid tool calling
• command-r7b-12-2024 — Lightweight, near-free

Strengths

• Native RAG / document grounding — Cohere models are built for retrieval-augmented generation
• Multi-step agentic tool use — Built into the model, not just bolted on
• Structured JSON output — Reliable JSON generation
• Complex reasoning chains — Excellent at multi-hop reasoning

Full Example

pythonfrom mtp import Agent
from mtp.providers import Cohere

Agent.load_dotenv_if_available()

provider = Cohere(
    model="command-a-03-2025",
    temperature=0.3,
    max_tokens=4096,
    preamble="You are a precise tool-using assistant.",
)

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

reply = agent.run_loop(
    "Calculate (25 * 4) + 10 and explain.",
    max_rounds=3,
)
print(reply)

Notes

• Uses Cohere V2 API (ClientV2) with OpenAI-compatible message format.
• Tool names with dots (.) are converted to double underscores (__) for Cohere compatibility, then converted back in responses.
• The preamble parameter is Cohere's equivalent of a system prompt. If a system message is already in the conversation, preamble is not injected.
• Text-only input (no image/audio/video/file support).

Source

src/mtp/providers/cohere_provider.py