Roo Code
also known as Roo Cline
Type: full-code · Vendor: Roo Code Inc. · Language: TypeScript · Status: discontinued · Status in practice: deprecated · Discontinued: 2026-05-15
Open-source VS Code AI coding agent (originally a fork of Cline) with role-based modes - Architect / Code / Ask / Debug / Orchestrator - plus MCP integration and custom-mode authoring. Extension archived 2026-05-15; vendor pivoted to Roomote (cloud agent on Slack).
Description. Roo Code was a VS Code extension that exposed a multi-role agent: Code for everyday edits, Architect for planning, Ask for explanations, Debug for tracing issues, and Orchestrator for delegating to other modes. Custom modes let teams add specialised roles. MCP servers extended the toolchain. The extension was shut down on 2026-05-15; the vendor (Roo Code Inc.) now ships Roomote, a Slack-integrated cloud engineering agent. Documentation directs former users to ZooCode or Cline as forks/alternatives.
Agent loop shape. Mode-switched tool-using loop inside VS Code. User picks (or Orchestrator picks) a mode; that mode's prompt+tool budget runs against the LLM; tool calls (file edits, exec, MCP) execute with optional approval; results return; Orchestrator may delegate further subtasks back to other modes.
Primary use cases
- role-typed VS Code agent (Architect / Code / Ask / Debug / Orchestrator)
- Orchestrator-mode delegation to other modes
- custom modes for team-specific workflows
- MCP-extended toolchain inside VS Code
Key concepts
- Modes (Architect / Code / Ask / Debug / Orchestrator) → role-typed-subagents (docs) — Role-typed prompts keep models focused on a single task.
- Orchestrator mode → supervisor (docs) — Delegates to other modes; recommended for complex multi-step tasks.
- Custom modes (docs) — Author specialised modes for team or workflow.
- MCP → mcp (docs) — MCP server integration for external tools.
Patterns this full-code implements —
- ✕Role-Typed Subagents
Architect / Code / Ask / Debug / Orchestrator are role-typed prompts.
- ★★Supervisor
Orchestrator (aka Boomerang) mode is the canonical supervisor primitive — breaks down complex tasks and delegates subtasks to specialised modes.
- ★★Model Context Protocol
MCP server support is listed as a core capability; per-tool 'Always allow' approval is built into the MCP integration.
- ★★Approval Queue
Per-tool approval flow with optional auto-approval; documented in Auto-Approving Actions with permission tiles per action category.
- ★Code-as-Action Agent
File edits and command execution are the primitive actions in Code/Debug modes; documented as native file system access and terminal control.
- ★★Tool Use
Mode prompts plus MCP tools form the tool surface; auto-approval permissions are configured per tool category (read, write, command, MCP).
- ★Subagent Isolation
Mode-as-subagent gives prompt+tool isolation: each mode (Code, Architect, Debug, Ask) has its own prompt and toolset; no full process-level isolation.
Neighbourhood
Click any neighbour to follow the lineage. Scroll to zoom, drag to pan.