agno-cursor-agent

Agno Model adapter that talks to the Cursor CLI over a persistent ACP (JSON-RPC) session.

Reasoning

Cursor unlocks multiple exciting features: multiple model providers, cost router, agentic memory, code indexing, cloud agent provisioning and a powerful harness. Now available to power your Agno agents for the best aSDLC pipeline experience.

Quickstart

Install from PyPI:

pip install agno-cursor-agent
# AgentOS with WebSockets:
pip install "agno-cursor-agent[standard]"

Import name on PyPI is agno-cursor-agent; in Python you still import agno_cursor.

Quick usage

from agno.agent import Agent
from agno_cursor import CursorACPChat, HarnessSettings

settings = HarnessSettings(acp_trace=False)
model = CursorACPChat(agent_key="coder", model="composer-2.5", mode="ask", settings=settings)

agent = Agent(id="coder", model=model, instructions=["You write code."])
# Requires cursor-agent login on the machine running this code.

Harness environment variables use the AGNO_CURSOR_ prefix (for example AGNO_CURSOR_ACP_TRACE=true).

Demo team

A full Agno AgentOS + A2A team lives under examples/demo_team. It uses CURSOR_TEAM_* settings and role-based models (leader defaults to auto).

cd examples/demo_team
pip install -e "../..[standard]"
cursor-agent login
python team_os.py

Connect the Agno UI to http://localhost:7777 (not 0.0.0.0).

Development

• Unit tests mock ACP I/O; CI never calls cursor-agent.
• Integration tests: pytest -m integration locally with login and optional team_os running.

See CONTRIBUTING.md.