What Is T3 Code? Setup, Features and Use Cases

TL;DR

• T3 Code brings Claude Code, Codex, Cursor, and other AI coding agents into one interface using your existing subscriptions.
• It lets you run multiple coding tasks, manage Git worktrees, review changes, and create pull requests.
• You can control agents from desktop, web, and mobile apps, or connect external agents through its MCP server.
• T3 Code is free and MIT-licensed, but it's still early software, so expect occasional bugs.

What is T3 Code?

T3 Code is an open-source control interface for AI coding agents, developed by T3 Tools. Its maintainers describe it as an agent harness control surface, meaning it lets developers interact with different coding agents through one application.

Unlike Claude Code or Codex, T3 Code isn't an AI model or coding agent itself. Instead, it runs and manages the agents you already use on your computer, with your existing subscriptions or API credentials.

For example, you could use Codex to implement a feature, switch to Claude Code for another task, and review both agents' changes without leaving T3 Code.

While a typical Claude Code GUI or the Codex app focuses on one provider's ecosystem, T3 Code brings multiple providers together.

The project is available on the official pingdotgg/t3code GitHub repository under the MIT license.

Why is T3 Code so popular?

Developers increasingly use multiple AI coding agents, but switching between their separate applications and terminals can complicate everyday workflows. T3 Code attempts to solve this problem with a unified, open-source interface.

The project has attracted substantial attention since its public release in March 2026. As of November 4, 2026, its GitHub repository had approximately 45,000 stars and 9,300 forks.

Several features help explain its appeal:

• Multiple AI providers. Developers can use their existing agent subscriptions rather than committing to one ecosystem.
• Parallel development. Separate threads and Git worktrees make it easier to supervise several coding tasks.
• Remote access. Desktop, browser, and mobile applications let developers manage agents running on another machine.
• Open-source code. T3 Code is free, MIT-licensed, and available to modify or fork.

The team says it built T3 Code after experimenting with existing interfaces, including the Codex app, Conductor, Claude Desktop, and Cursor Glass.

Its functionality has also expanded considerably since launch. What began as a Codex-focused project now supports additional coding agents, remote workflows, and mobile applications.

This growth reflects broader interest in tools that make working with AI coding agents more manageable, including projects such as REA (Reverse Engineer Anything).

Key features of T3 Code

T3 Code combines several tools for managing AI-assisted development, from running coding agents to reviewing their changes and accessing them remotely.

One app for multiple coding agents

As of November 2026, T3 Code's main README lists support for Claude Code, Codex, Cursor, Grok Build, OpenCode, Google Antigravity, and fx.

Each provider uses its own authentication and access arrangements. You don't buy AI models through T3 Code or receive an additional usage allowance. Instead, the application works with the subscriptions or API access you already have.

T3 Code also supports separate account configurations for providers such as Claude Code and Codex.

This makes the application particularly useful for developers who regularly switch between agents or want to compare their approaches to the same problem.

Threads, Git worktrees, and pull requests

T3 Code organizes coding tasks into threads. Each thread can run in your existing project directory or an isolated Git worktree.

A worktree lets you work on another branch in a separate directory without changing your main checkout. This means multiple agents can work on different tasks without interfering with each other's files.

For example, one agent could implement a feature while another fixes a bug in a separate worktree.

T3 Code also provides interfaces for reviewing diffs, managing commits, and creating pull requests. Its source-control integrations support GitHub, GitLab, Bitbucket, and several other Git hosting services.

Permission modes

T3 Code provides different permission modes to control how freely an agent can modify files or execute commands.

The available modes include:

• Supervised. Requests approval for commands and file changes.
• Auto-accept edits. Approves file edits automatically, while other actions may still need approval.
• Auto. Uses the provider's automatic review system where available.
• Full access. Allows supported actions without routine approval prompts.

The available behaviour depends on the coding agent, and some providers impose additional restrictions.

These settings help developers balance convenience and oversight when running automated coding tasks.

Desktop, web, and mobile apps

The T3 Code mobile app is one of the project's more distinctive features. It lets developers supervise coding agents from an iPhone or Android device while the actual work continues on another computer.

You could start a task at your desk, check its progress while away, and review the agent's output from your phone.

T3 Code supports several interfaces:

For remote connections, T3 Code offers its own T3 Connect service, local network pairing, and options such as Tailscale HTTPS.

Importantly, the machine running your agents must remain powered on and reachable. The mobile application acts as a remote interface rather than running the coding agents directly.

An MCP server for external agents

T3 Code also includes a Model Context Protocol (MCP) server, which lets compatible external agents interact with a T3 Code environment.

For example, an agent running separately in Claude Code, Codex, or ChatGPT can connect to T3 Code to inspect projects and threads.

With additional permissions, external agents can also start threads, send messages, and coordinate work.

Connections require authentication, and read-only access is the default. Developers can grant broader permissions when approving a connection.

See our guide on how MCP servers are set up for more background, or consult T3 Code's official MCP documentation.

How to set up T3 Code

Getting started with T3 Code is relatively straightforward, especially if you already use a supported AI coding agent.

Step 1. Prepare your coding agent

T3 Code doesn't replace or automatically install the underlying coding agents.

Before starting, make sure you have access to at least one supported provider, such as Claude Code, Codex, or Cursor.

Install and authenticate the provider on the computer where T3 Code will run. The official provider documentation explains the requirements for each agent.

Step 2. Install T3 Code

For macOS and Linux, the official command-line installer is:

curl -fsSL https://t3.codes/install.sh | sh

Alternatively, you can download the desktop application from the T3 Code GitHub Releases page.

The official T3 Code download options include Homebrew for macOS, winget and Scoop for Windows, and packages for several Linux distributions.

The repository also provides T3 Code nightly builds for developers who want to try the latest changes. These may introduce new features sooner but can be less stable than regular releases.

For a one-time test without installing T3 Code globally, the README also documents npx t3@latest.

Step 3. Start T3 Code and open a project

After installing the CLI, run t3 to start the local server and open its web interface. Alternatively, launch the desktop app.

Select or add a project directory, choose your coding agent, and create a new thread.

You can then describe a task, such as implementing a UI component, writing tests, or investigating a bug.

For separate development branches, create a thread in a new Git worktree.

Step 4. Connect another device (optional)

To access your coding environment remotely, install the T3 Code mobile app or open the web interface on another device.

T3 Connect provides a way to connect without manually configuring router forwarding. Local network pairing and Tailscale are also available.

For detailed instructions, consult the official installation and remote-access guides.

T3 Code use cases

T3 Code is particularly useful when developers want to supervise several AI coding tasks or work with coding agents across different devices.

Run multiple coding agents in parallel

Suppose you're developing a web application and need to implement authentication, improve the interface, and fix existing tests.

You could assign these tasks to separate threads and Git worktrees, allowing agents to work independently.

Afterward, you can review the changes and decide which branches to merge.

Compare coding agents on the same task

Different coding agents can approach the same problem differently.

T3 Code lets you start separate threads using different providers, making it easier to compare their implementations.

For instance, you could give Claude Code and Codex the same feature request, review their solutions, and decide which one to keep.

Supervise development from your phone

T3 Code's remote functionality is useful when an agent continues working after you've stepped away from your computer.

You can monitor progress, respond to prompts, and review results from a mobile device without needing to return to your desk.

This is especially useful for longer development tasks that require occasional supervision.

Coordinate agents through MCP

T3 Code's MCP server lets other compatible agents interact with its projects and threads.

An external agent could inspect existing tasks, create new threads with the necessary permissions, or check their progress.

This allows T3 Code to participate in larger automated development workflows instead of operating solely as a graphical interface.

Extend coding agents with external tools

Coding agents can also benefit from additional tools that provide access to external information.

For example, a developer building a market research application might need current website content or search results.

MCP-compatible services can provide this information directly to supported coding agents, reducing the need to build separate data retrieval components.

See our overview of MCP servers for Cursor or web scraping CLIs for AI agents for potential additions to an agent's workflow.

T3 Code vs. Claude Code, Codex and other agent apps

T3 Code vs. Codex isn't a direct comparison because Codex is an AI coding agent, while T3 Code is an interface that can run it.

The same applies to Claude Code. A more useful comparison involves the applications developers use to manage these agents, such as the Codex app or Conductor.

T3 Code stands out for its open-source code, broad platform availability, and support for several coding providers.

Claude Code and the Codex app may be more convenient for developers committed to one ecosystem, while Conductor offers another multi-agent workspace, particularly for Mac users.

Other open-source T3 Code alternatives include Orca and Paseo, both of which help developers manage multiple AI coding agents.

The best choice ultimately depends on your preferred agents, operating system, and need for remote access or parallel development.

Limitations to know before you switch

T3 Code is a promising project, but there are several things worth considering before making it your primary development environment.

First, the maintainers explicitly describe it as early software and warn users to expect bugs. Its rapid development also means features, supported providers, and documentation can change frequently.

Second, T3 Code depends on the coding agents you connect. It doesn't provide its own language model or replace the subscriptions and API credentials those agents require.

Third, although T3 Code runs its environment on your machine, your prompts and relevant code context still go to the underlying AI providers. Review their data-handling policies when working with sensitive projects.

Finally, the project's documentation currently lives inside its GitHub repository, and the team isn't broadly accepting major contributions.

These limitations don't necessarily outweigh the benefits, but they're worth keeping in mind if you depend on AI coding agents for everyday development.

Track trending repos and give coding agents live web data

Discover projects like T3 Code with GitHub Research

Want to find more fast-growing developer tools before they become widely known?

Decodo GitHub Research is an open-source application for discovering trending and emerging GitHub repositories, researching projects by keyword, and generating AI-powered reports.

It offers three research modes: Trending, Emerging, and Keyword. You can also save recurring monitors, compare repository activity over time, and export findings as Markdown or JSON.

For example, you could research keywords such as coding agent or MCP to discover projects related to T3 Code.

GitHub Research retrieves public repository information through Decodo's Web Data API and uses your chosen LLM provider to analyze the results.

Give your coding agents access to live web data

T3 Code manages coding agents, but applications those agents build may also require fresh information from the web.

The Decodo MCP Server provides MCP-compatible agents with tools for retrieving public website content, search engine results, and structured data from supported platforms.

For example, an AI coding agent building a price-monitoring application could use these capabilities to retrieve current product information without implementing its own web scraping infrastructure.

Decodo handles tasks such as JavaScript rendering and website unblocking through its managed service.

You can configure Decodo MCP Server with supported coding agents such as Claude Code or Cursor. The integration belongs to the underlying agent, rather than being a native T3 Code feature.

For further options, see our overview of MCP servers for AI workflows.

Bottom line

T3 Code is worth considering if you regularly use multiple AI coding agents, want to manage parallel development tasks, or need to supervise your agents from another device.

Its open-source nature and broad provider support make it an attractive option, although its early development stage means you should expect occasional issues.