Cline vs Roo Code: What Happened and What to Use

Roo Code is gone. Its docs say the extension was shut down on 15 May 2026, and its repository is archived. The two live options are Cline, the project Roo Code started from, and Zoo Code, a community fork that continues it. Choose Cline for the actively released original with a CLI, desktop app and JetBrains plugin. Choose Zoo Code if you want the Roo modes and settings to carry on. Checked 31 October 2026.

The short version

If you are here because an older guide told you to install Roo Code, do not. The marketplace listing and the docs both say it is shut down.

What happened to Roo Code

The Roo Code documentation opens with a notice: the Roo Code Extension was shut down on May 15th. If you want an alternative, it points to ZooCode, a fork started by the Roo Code community, and to Cline, from which Roo Code originated. For paying users with billing questions it gives a billing address. The GitHub repository RooCodeInc/Roo-Code is archived, with its last release, v3.54.0, dated 15 May 2026 and about 24,000 stars.

The Zoo Code README explains the other side: Zoo Code continues development after the Roo team wound down active Roo Code work to focus on a different product called Roomote, and it thanks the Roo team. The core Zoo team is a group of developers who contributed to Roo before. The repository Zoo-Code-Org/Zoo-Code is Apache-2.0, was created on 23 April 2026, has about 2,100 stars and was pushed to on 30 October 2026. It publishes a Roo to Zoo migration guide.

So the choice is no longer Cline against Roo. It is Cline against Zoo, with Roo as history. Our entries are Cline and Roo Code.

What Cline is now

Cline's README calls it the open source coding agent in your IDE, terminal and desktop. The repository is Apache-2.0, has about 71,000 stars and had its latest release, v4.1.26, on 26 October 2026. It ships in several forms:

• CLI. Install with npm i -g cline; runs interactively or headless.
• Desktop app. A native app for macOS and Windows that manages models, plugins and MCP servers, plus routines.
• VS Code extension. Creates files, runs commands, browses the web and uses tools with human-in-the-loop approval.
• JetBrains plugin.
• SDK. Build your own agents on the same engine, with custom tools, multi-agent teams, connectors and scheduled automations.

Features the README lists: Plan and Act modes (explore and agree a strategy, then execute), rules in .clinerules files, skills, MCP servers, plugins with lifecycle hooks for logging and policy enforcement, multi-agent teams where a coordinator delegates to specialists, and watching of linter and compiler errors as it works. Providers include Anthropic, OpenAI, Ollama and LM Studio for local models, and any OpenAI-compatible API.

What Zoo Code is

Zoo Code describes itself with Roo's old tagline, a whole dev team of AI agents in your code editor. It is a VS Code extension, listed on the Marketplace under ZooCodeOrganization. Because it continues Roo, it carries Roo's feature set: modes such as Code, Architect, Ask and Debug, custom modes for your team's workflow, and MCP server support, per the Roo README that it descends from.

Its backing is a community team rather than a company, which is both an advantage (no corporate pivot) and a risk (funding and bus factor). It is young: the repository is about six months old.

Side-by-side facts

Cost

Cline's pricing page says it is free for individual developers: you pay only for AI inference on a usage basis, with no subscription and no vendor lock-in. You can bring your own keys from providers such as OpenAI, Anthropic and Google, or use the Cline provider. There are no subscription or seat fees for the open source version. A separate Enterprise offering exists for larger organisations.

Zoo Code, as a community fork of an open-source tool, is expected to follow the same bring-your-own-key model.

In both cases, the real cost is the model. A coding agent sends your files and a long conversation to the model on every step, so a hard task costs more than a quick one. Both docs lean toward quality over thrift: Roo's own docs described its approach as trading tokens for quality. If you want to cap spend, use a cheap model for simple tasks. See DeepSeek vs Claude for the price gap, and Ollama vs LM Studio for running a model locally at no per-token cost.

How they work day to day

Cline. You give a task. In Plan mode it reads the codebase, asks questions and proposes a strategy. In Act mode it edits files and runs commands, asking your approval for each step unless you auto-approve. It watches terminal output as commands run and reacts to linter and compiler errors. Rules in .clinerules shape its habits. Through the CLI you can run it headless, for example with a team name and a task in one command.

Zoo Code. The familiar Roo flow: pick a mode for the job. Architect to plan, Code to implement, Debug to chase a failure, Ask to question the codebase. Custom modes let a team encode its own workflow, such as a reviewer with read-only tools.

The philosophies differ a little. Cline's Plan and Act is one switch with two states. The Roo family has more modes with narrower roles. If you like specialised roles, you will like the Roo lineage. If you want fewer settings, Cline is simpler.

Migrating from Roo Code

If you used Roo Code, you have two paths.

To Zoo Code. Follow the Roo to Zoo migration guide linked from the Zoo README. It is the shortest path since the product descends from Roo, and your modes and habits should transfer. Check which settings carry over.

To Cline. Install the extension or CLI, add your provider keys, and rewrite your custom modes as .clinerules and skills. Re-add your MCP servers through Cline's settings. Expect a day of adjustment.

In both cases, uninstall the dead Roo extension, so it does not conflict, and keep a copy of your custom mode definitions before you do.

MCP and add-ons

Both lines support MCP servers, which is how an agent reaches GitHub, databases and browsers. Cline's README says to use MCP to connect to databases, query APIs and manage cloud infrastructure, and it links community-built servers. The desktop app manages them in its settings. For starting points, see best MCP servers, best browser MCP servers and best Git MCP servers.

Privacy and data handling

Both are extensions that run locally and send prompts to whichever model provider you configure. Their privacy depends on that provider's terms, and on any hosted provider the tool offers. With a local model through Ollama or LM Studio, nothing leaves your machine.. Cline's approval step is a real control: it asks before it runs commands, so read each one.

Limits and gotchas

• Do not install Roo Code. Its marketplace listing and docs say it is shut down.
• Zoo Code is young. About six months old, with a community team. Watch its release rhythm.
• Costs scale with tasks. Agents resend context at every step.
• Auto-approve is risky. Turn it on only in a sandbox or a disposable branch.
• Different VS Code forks. Check which marketplace each extension is on if you use a VS Code fork rather than VS Code itself.
• Names are similar. Cline, Roo, Zoo and Roomote are four different things. Roomote is a separate product from the Roo team.

Who should pick which

A new user

Cline. It is the original, actively released, with the widest set of surfaces.

A Roo Code user

Zoo Code first. If it fits, stay. If you want a company-backed project or a CLI, move to Cline.

A JetBrains or terminal user

Cline, the only one of the two with those surfaces.

A team that wants audit hooks

Cline's plugin system describes lifecycle hooks for logging, auditing and policy enforcement. Test them against your policy.

Someone choosing between agents in general

Compare with Aider vs Claude Code, Cursor vs Windsurf and the open-source agents list.

A first hour with Cline

1. Install the extension from the VS Code Marketplace, or the CLI with npm i -g cline, or download the desktop app for macOS or Windows.
2. Add a provider. Bring your own key from Anthropic, OpenAI or Google, point it at an OpenAI-compatible endpoint, or run a local model with Ollama or LM Studio.
3. Open a repository and write a .clinerules file with three lines: how to run tests, which folders are off limits and the commit style.
4. Start in Plan mode with a small task, such as adding a validation rule and a test. Read the plan before you switch to Act.
5. Keep approvals on. Read each command Cline proposes before you accept it.
6. When the task ends, read the diff as you would a colleague's pull request.
7. Check what the session cost on your provider's dashboard, then decide whether a cheaper model could have done the job.

Repeat the same task in Zoo Code if you are weighing the two. Use the same model and the same prompt, so the difference you see is the tool.

A note on picking tools in a fast market

This pair is a reminder that coding-agent tools move quickly, and that open-source ones can change hands. A few habits protect you. Keep project instructions in plain files in your repository, such as an AGENTS.md or .clinerules, so a new tool can read them. Keep MCP server configuration in a form you can copy. Prefer tools with permissive licences, since a fork can keep them alive. And do not build a team workflow on one maintainer's goodwill without a plan to move.

Our take

The story here matters more than the feature table. An open-source agent that many people used was wound down by its team in a few months, and a community fork picked it up. That is a good reason to prefer tools whose licence lets you fork, and to keep your rules and prompts in files you control rather than in a product's settings.

For most people, Cline is the safe pick. It is active, broad and Apache-2.0. Zoo Code is the natural pick for a Roo user, and worth watching. We did not test either on a shared task, so we make no quality claims. The facts come from the repositories and docs on 31 October 2026, and open points are marked.

Related pages

• Cline and Roo Code
• Open-source AI agents
• Claude Code vs OpenCode