SuperDev Docs

SuperDev Docs

SuperDev gives AI coding agents runtime hands and eyes. The point is not to teach you how to click every panel. The point is to give the agent the right capabilities, rules, and approval boundary so it can operate services, collect evidence, and prove the result.

You Are Not Here To Click Panels

The Agent-first workflow is:

1. You provide the goal, entrypoint information, and approval boundary.
2. The agent uses SuperDev MCP tools to inspect, model, run, debug, and deploy.
3. The agent generates config diffs; real write operations wait for your approval when policy requires it.
4. The agent reports runtime evidence instead of saying only that code changed.

Start By What You Want The Agent To Own

Tier	You want the agent to	You provide / confirm	The agent owns
1. Local development	Code and debug locally	Repository entrypoint, local browser-debug boundary, optional debug guidance and credentials	Model services, generate runtime/deployment config, start/restart services, read logs, use browser tools, capture breakpoints
2. Remote debugging	Investigate remote or production-like bugs	Host, remote superdev-agent install method, remote service and log hints	Generate remote deployment / log source config, aggregate remote logs, run diagnostics, collect evidence, explain root cause
3. Team deployment	Deploy and roll back	Pipeline config; humans still configure ingress, SSL, and DNS	Validate pipelines, deploy, roll back, read run logs, manage artifacts

Connect Once

Start with Connect an AI agent. SuperDev supports Claude Code, Codex, and Cursor, installs the MCP entry, and installs the SuperDev skill that teaches the agent how to use the tools safely.

Keep The Standard

Useful agents do not stop at “changed the code.” They prove the result with service status, log context, browser screenshots, console/network output, debug stack frames, pipeline run logs, artifacts, or tests. See Self-healing loop.