← Docs

Agents

Sideboard shells out to each agent’s own CLI. Install the ones you want on your PATH and authenticate. Cursor uses a user-installed @cursor/sdk plus a CURSOR_API_KEY (Settings → Agents → Cursor → Install). Sideboard does not pack the Cursor SDK — so agents can update it without a new Sideboard build.

npm i -g @anthropic-ai/claude-code
npm i -g @openai/codex
npm i -g opencode-ai@latest
npm i -g @cursor/sdk
export CURSOR_API_KEY="cursor_..."

Then sideboard detect (or Settings → Agents) shows which harnesses are installed and signed in.

Agent How it runs
Claude Code Official CLI on PATH
Codex Official CLI on PATH
OpenCode Official CLI on PATH
Cursor User-installed @cursor/sdk + API key
Brightsy Optional chat-only adapter

Mechanical control (list, send, diff, land) stays on the Sideboard CLI — zero tokens. Use MCP when an agent needs judgment across threads or needs to open a pane in the desktop.

Connect the Sideboard MCP from each harness so an agent can drive the board. Typical stdio command: sideboard mcp (or npx -y sideboard-mcp). Packaged Sideboard.app writes the absolute Node + bundled MCP path on launch.

A Claude worktree turn also loads MCP servers from plugins enabled for that checkout (enabledPlugins in ~/.claude/settings.json, then the project’s .claude/settings.json, then .claude/settings.local.json). A server already in ~/.claude.json wins on the same name. Orchestration turns stay on the injected Sideboard server. Plugin UIs render in the MCP App frame on the desktop and the phone.

See Remote integrations for Settings → Agents vs Git / Issues / Remote / Connectors.