Your agents can read and write code. They just can't see or interact with what they build. sidekar gives them browser control, desktop automation, and a way to talk to each other.
curl -fsSL sidekar.dev/install.sh | sh
copy
copied
Coding agents generate code blind. They write a React component but can't see if it renders. They update a form but can't test if it submits. sidekar bridges the gap between writing code and verifying it works — giving agents the same feedback loop a developer gets from their browser.
Four capabilities, one tool server.
Full Chrome DevTools Protocol access. Navigate, click, type, screenshot, read page content. Smart perception with auto-brief and escalation — not raw DOM dumps that burn tokens.
macOS Accessibility API integration. Find and interact with native app elements, take screenshots, launch and manage applications. No browser required.
Multiple agents share one Chrome instance with tab isolation. Each agent owns its tabs. No conflicts, no cross-contamination, full concurrency.
Agents discover and message each other across tmux sessions. Send requests, broadcast, hand off tasks. The coordination layer for multi-agent workflows.
One command downloads the binary and auto-configures every MCP client it finds on your machine. Claude Code, Cursor, Copilot, Codex, Gemini CLI — all set up in seconds.
Ask your agent to test a UI, check a deployment, fill out a form, or research something on the web. It launches Chrome and acts, just like you would.
Smart perception returns compact page summaries, not entire DOMs. Agents escalate from text to screenshot only when needed. Your tokens go to thinking, not parsing.
Simple, honest, honor-system.
If your company pays you to write code, it's professional use.