How we set up our cloud agent environment
Source: https://cursor.com/blog/cloud-agent-environment
Author: Mathew Hogan (Cursor)
Published: 2026-07-30
Summary
Cursor shares how they built the development environment that enables cloud agents to author over 50% of merged PRs to their monorepo, up from roughly 1 in 10 in December 2025.
Three pillars:
1. Cloud matches local development: Mac local → Ubuntu cloud VM migration via Dockerfile; security features including network egress restrictions, scoped/proxied git, secret scanning in commits, secret redaction in tool results.
2. anydev CLI: Reduced "command entropy" — agents previously struggled with convoluted build commands and footguns. anydev provides a unified interface for starting services, with help menus and a supervisor process that monitors long-running builds.
3. Cloud Doctor (self-healing environment): An automation built on Cursor Cloud MCP that periodically checks for failures, distinguishes transient vs. salient errors, performs root cause analysis, and opens PRs to fix issues with high confidence. Cloud Doctor agents also inspect other agents' traces to find misleading skills or slow workflows, then fix them.
Key metrics:
- Cloud agents: ~10% → 56%+ of merged PRs (7-month period)
- Agents can now share recorded demos and, for many tasks, engineers merge without checking out locally
Three readiness questions:
1. Do agents have access to the same tools and data a developer would? 2. Can agents find skills that document how your developers actually work? 3. Can agents test and verify core workflows?
Evidence (Obsidian)
From 调研 2026-07-31:
56% 是环境产品做出来的组织吞吐样本，不是“模型更会补全”的金句。采纳时必须并列 revert rate、人审工时、MTTR。把「agent 专用 DX」写进团队就绪清单；merged% 单独出现时默认不完整。