Roo Code in 2026: status, modes, custom modes, rules, MCP and alternatives

Is Roo Code still available?

Roo Code, the open source AI coding agent for VS Code, was shut down on May 15, 2026. The Roo Code repository is archived on GitHub, and its README now says "The Roo Code Extension was shut down on May 15th." The last release was v3.54.0, published the same day.

The shutdown covered more than the extension. Roo Code's sunset notice says "All Roo Code products will be shut down on May 15, 2026," including Roo Code Cloud and Roo Code Router, with unused balances refunded. The team's next project is Roomote at roomote.dev, and roocode.com now redirects there.

The community did not stop there. Zoo Code, a fork started by former Roo contributors, says it "continues development of this project after the Roo team wound down active Roo Code work to focus on Roomote." It ships regular releases (v3.92.0 came out on October 24, 2026).

This page explains what Roo Code was, how its main features worked, and what your options are if you still have a Roo Code setup.

What was Roo Code?

Roo Code's docs described it as "the autonomous AI coding agent that helps you code faster and smarter in VS Code," and the extension was free and open source. It ran in your editor, could read and write files and run terminal commands with your approval, and worked with many model providers instead of one. Its docs said it passed 3 million extension downloads.

It began life as Roo Cline. The changelog shows early releases such as "Roo Cline now can run side-by-side with Cline" and "Incorporate MCP changes from Cline 2.2.0," and version 3.2.0 announced the "Name Change From Roo Cline to Roo Code." The same release introduced custom modes, which became Roo Code's defining feature. For the comparison with its parent, see Cline vs Roo Code.

How did Roo Code modes work?

A mode is a persona with its own role, instructions and tool access. The modes docs list five built-in modes:

You switched modes from the dropdown by the chat box, with slash commands such as /architect or /debug, or with Cmd+. (Ctrl+. on Windows and Linux) to cycle through them. Each mode remembered the last model you used with it, so you could plan with one model and code with another.

How did custom modes work?

Custom modes let you define your own personas. Project modes lived in a .roomodes file at the repo root (YAML or JSON), and global modes in custom_modes.yaml, opened from the Modes page with "Edit Global Modes." A YAML example from the docs:

customModes:
  - slug: docs-writer
    name: 📝 Documentation Writer
    description: A specialized mode for writing and editing technical documentation.
    roleDefinition: You are a technical writer specializing in clear documentation.
    whenToUse: Use this mode for writing and editing documentation.
    customInstructions: Focus on clarity and completeness in documentation.
    groups:
      - read
      - - edit
        - fileRegex: \.(md|mdx)$
          description: Markdown files only

The fields that mattered most:

• roleDefinition went at the start of the system prompt and set the mode's identity.
• whenToUse told the Orchestrator and mode switching when to pick this mode.
• groups set the tool groups the mode could use, and fileRegex limited which files the edit group could touch.
• customInstructions added rules near the end of the system prompt.

Modes could be exported to a YAML file with their rules and imported at project or global level, which made them easy to share across a team.

How did Roo Code rules and AGENTS.md work?

The custom instructions docs describe a folder-first system:

• Workspace rules: every file in .roo/rules/, with a single .roorules file as a fallback when that folder is empty or missing.
• Mode rules: .roo/rules-{modeSlug}/, such as .roo/rules-code/, with .roorules-{modeSlug} as the fallback.
• Global rules: ~/.roo/rules/ and ~/.roo/rules-{modeSlug}/, loaded first, with workspace rules winning on conflicts.
• Legacy files: .roorules and .clinerules at the root, used only when no rules folder had content.
• AGENTS.md: loaded from the workspace root by default (AGENT.md as a fallback), controlled by the roo-cline.useAgentRules setting.

If you are moving off Roo Code, AGENTS.md is the file most worth keeping, because most current agents read it.

Did Roo Code support skills and MCP?

Skills. Yes. Roo Code read SKILL.md folders from .roo/skills/ and .agents/skills/ in a project, and ~/.roo/skills/ and ~/.agents/skills/ globally. It also supported mode-specific folders such as .roo/skills-code/. The skills docs required the name to match the folder, 1 to 64 lowercase letters, numbers and hyphens, and a description of 1 to 1,024 characters, which matches the Agent Skills standard. Skills in .agents/skills/ move to other agents without changes.

MCP. Yes. The MCP docs describe a global mcp_settings.json and a project .roo/mcp.json that you could commit. Both used an mcpServers object, with optional alwaysAllow, disabled and disabledTools fields per server. It supported stdio, Streamable HTTP ("type": "streamable-http") and legacy SSE servers.

Providers. Roo Code was model-agnostic: Anthropic, OpenAI, Gemini, Amazon Bedrock, Vertex, OpenRouter, DeepSeek, Mistral, xAI, LiteLLM, any OpenAI-compatible endpoint, and local models through Ollama or LM Studio. Roo Code Router, its own key-free option, closed with the shutdown.

Can you still install and use Roo Code?

The installing docs told you to search for "Roo Code" by RooVeterinaryInc in the VS Code Marketplace, or in Open VSX for editors like VSCodium. As of 28 October 2026, the listing still appears at v3.54.0. It works with your own API key, but it will not get new model support, bug fixes or security patches, and the Apache 2.0 license comes with Roo Code, Inc.'s "AS IS" disclaimer in the README.

For most people, moving is the better choice:

• Zoo Code keeps the Roo Code modes and adds its own features, such as per-mode MCP restrictions and a Destructive Command Guard. Its migration guide is two steps: export settings in Roo Code, then click Import Settings in Zoo Code. The exported file may contain API keys, so store it carefully.
• Cline is the project Roo Code came from, and Roo's sunset notice recommends it, saying "They've incorporated much of what we built." It uses Plan and Act modes instead of custom modes. See the Cline guide.
• Other open source agents such as Kilo Code are worth a look if you want a VS Code agent with your own keys.

How do you keep your Roo Code setup when you switch?

Your rules, modes and skills are plain files, so most of the work carries over:

1. Copy the contents of .roo/rules/ into an AGENTS.md at the repo root.
2. Move skills from .roo/skills/ to .agents/skills/, which several agents read.
3. Copy server entries from .roo/mcp.json into the new tool's MCP config.
4. Turn the role and instructions of each custom mode into a rule or skill if the new tool has no modes.

If those files are scattered across repos, Kipt keeps prompts and skills as Markdown in a private GitHub repo you own, and can import skills from a GitHub link, a folder or a .zip. That gives you one copy to install into whichever agent you pick next.