How to install and use Roo Code in VS Code: model and first task

Install the Roo Code extension, connect a provider model, then land one reviewable agent change in your own repo.

The first hour with Roo Code usually breaks on the wrong extension, a missing model key, or a first task with no reviewable diff. Install the Marketplace build from publisher RooVeterinaryInc, connect a provider in settings, then change one reversible line in a repo you own.

The query is how to install and use Roo Code in VS Code. This page follows archival install and provider notes at docs.roocode.com, roocodeinc.github.io/Roo-Code, and Marketplace item rooveterinaryinc.roo-cline. It is a VS Code agent extension, not this site’s meeting product. To walk a diff on a short call, open a room on wbmeet; see how to create and join a room.

Install the extensionConnect a modelOne agent task

Check what you are installing

Roo Code is an open-source VS Code coding agent: it reads and writes files, can run terminal commands, and can use MCP tools. The extension is free. Inference uses your provider key (OpenRouter, Anthropic, OpenAI, Gemini, DeepSeek, and others) or a local Ollama / LM Studio model. Official docs require VS Code 1.84.0 or later. Publisher: RooVeterinaryInc. Extension id: rooveterinaryinc.roo-cline.

VS Code
Lives inside VS Code you already have
1 key
Provider API key or local model
4 modes
Code / Architect / Ask / Debug

First hour: install, connect a model, land a reviewable change

  1. 1

    Install from the Marketplace

    Open VS Code. Cmd+Shift+X (macOS) or Ctrl+Shift+X (Windows/Linux). Search Roo Code, pick publisher RooVeterinaryInc, Install. Reload if asked. If Marketplace is gone, use Open VSX or a .vsix from GitHub RooCodeInc/Roo-Code Releases (Extensions → … → Install from VSIX). Skip unofficial mirrors.

  2. 2

    Connect a model in settings

    Open the Roo panel from the Activity Bar, then the gear. Choose an API Provider (OpenRouter, Anthropic, OpenAI, Google Gemini, DeepSeek are common). Paste the key into that provider’s field and pick a model. For offline use, choose Ollama or LM Studio. Keys go into VS Code Secret Storage. Without a key, chat cannot send.

  3. 3

    Open a repo you own

    File → Open Folder on a project you can roll back. Do not use a production secrets directory for the first experiment.

  4. 4

    Open chat and give a reviewable task

    Set the dropdown to Code (Architect to plan, Ask to explain, Debug to isolate; or switch with /). Ask what the project does and where the entry point is, then: “Add a unit test for one untested helper, or input checks on one public function. Show the diff and wait.” Mention files with @. Approve each tool use. Do not paste secrets. Do not turn on blanket auto-approve for commands.

# First hour · Roo Code in VS Code · as of 2026-09
# 1. Extensions → search Roo Code (publisher RooVeterinaryInc)
#    id: rooveterinaryinc.roo-cline   VS Code 1.84.0+
# 2. Activity Bar → Roo panel → gear → API Provider
#    Paste a key (OpenRouter / Anthropic / OpenAI / Gemini / …)
#    or Ollama / LM Studio for a local model
# 3. Open a repo you own. Code mode. Chat:
#    "Add a unit test for one untested helper; show the diff and wait."
# Approve each tool use. Do not paste secrets.

Install paths, providers, pitfalls

PathWhat official docs sayFirst-hour note
MarketplaceRooVeterinaryInc / Marketplace or Open VSXCheck the publisher name
VSIXGitHub Releases .vsixUse the last archive if listing is gone
Connect a modelGear → API Provider + keyExtension free; inference billed by provider
ModesCode / Architect / Ask / DebugConfirm Code before file edits
Tool useFile and terminal actions need approvalAuto-approve commands is risky
01

Panel stays silent

Usually a bad or expired key, or the provider is down. Recheck the key, network, View → Output → Roo Code, then restart VS Code.

02

Marketplace search is empty

Official shipping stopped. Try Open VSX, the last VSIX, or move to ZooCode / Cline. Need VS Code 1.84.0+.

03

First task stayed a chat

“What does this repo do” is warmup. Finish with a reviewable change and read the diff before you approve.

Archival install notes: search Roo Code (RooVeterinaryInc), reload, open the Activity Bar panel. Provider pages: gear → API Provider, paste a key, pick a model. Commands and prices follow the live provider docs.

Free extension ≠ free model
The extension is open source. Cloud providers charge per token. Ollama / LM Studio can stay off that bill.
Approve each step
The agent can read files, edit files, and run the terminal. Default is to ask first. Auto-approving commands is high risk.
This is not a meeting-product guide
Only the first hour of Roo Code. Short whiteboard call: wbmeet or when not to open another heavy meeting app.

Questions worth checking

Do I have to pay to run the agent?

The extension is free. Cloud models usually need a provider account and key, billed by usage. Ollama / LM Studio can be tried offline; quality follows the model you pull.

Can I still install after the official shutdown?

Yes, the last build (Marketplace if still listed, or a GitHub Release VSIX). It will not receive official updates. New work lives on ZooCode or Cline — follow those docs.

Where does the API key go?

Roo panel gear → API Provider. Choose OpenRouter, Anthropic, OpenAI, Gemini, or another listed provider, paste the matching key, pick a model. Keys sit in Secret Storage. Do not commit them.

Must the first task touch production code?

No. Ask what the project does, then one reversible change (test or validation) and read the diff first. Reject tool uses you do not want.

Start a free meeting