What is mpai? Features, Pricing & Tutorial (2026)

The Problem mpai Solves

Modern terminal AI coding agents like Claude Code and Codex have made individual developers remarkably fast, but they have also introduced a new kind of isolation. When a cofounder or teammate is hours deep into a terminal session debugging a complex architecture problem, that valuable context remains trapped on their local machine. Traditionally, bridging that gap meant resorting to cumbersome screen sharing, messy transcript copy-pasting, or asking the primary user to re-explain dozens of previous turns.

This friction disproportionately hurts pair programmers, cofounders, and engineering teams who want to share insights without interrupting their existing command-line workflows. Asking a teammate to abandon their preferred terminal setup just to collaborate creates unnecessary overhead.

mpai fixes this by acting as a lightweight coordination layer directly above the agents. It allows a named teammate to securely jump into an exact, ongoing terminal AI session with their individual identity intact, all while keeping the host machine in full control.

In this tutorial, you'll learn exactly how to use mpai — step by step.

How to Get Started with mpai in 5 Minutes

1. Ensure your host environment is running macOS and is connected to an active Tailscale private network.
2. Install mpai on the host machine using Homebrew via brew install godfaddaai/tap/mpai && mpai start or via the provided Node.js package installation command.
3. Follow the interactive prompts in the terminal to verify your machine, enter host and teammate names, and select an existing Codex or Claude Code session to share.
4. Copy the secure npx command generated by the tool and share it privately with your teammate.
5. Have your teammate paste the command into their terminal to join the exact session instantly with their identity attached.

How to Use mpai: Complete Tutorial

Step 1: Preparing Your Environment and Installing the Tool

Before launching mpai, verify that both your host machine and your teammate's client machine meet the system requirements. You will need macOS on the host, Node.js 20+ available, and an established Tailscale network connecting both devices. Open your primary terminal where you normally run Codex or Claude Code and ensure you have an active session running that you want to share.

Run the installation command on your host machine to pull the latest open-source build. The installation process sets up the local coordination dependencies without modifying your core agent installations. Once installed, invoke the start command to initialize the discovery process for your active terminal sessions.

Step 2: Discovering and Selecting Native Sessions

Once mpai launches, it automatically scans your local environment to discover native Codex and Claude Code sessions currently in motion. You do not need to export transcripts or move your work into a web browser or specialized IDE. The tool lists active sessions directly in your command-line interface for easy review.

Navigate through the detected sessions and choose the exact context you want to open up for collaboration. The tool prompts you to designate names and configure the boundary of the invite. By scoping the invite to a single chosen session, you ensure that unauthorized sessions on your machine remain entirely private and inaccessible.

Step 3: Generating Secure Invites and Sharing Context

After you select the target session, mpai generates a private, cryptographically secure connection string designed to be executed via npx. This command utilizes your private Tailscale tailnet to route traffic securely between machines without relying on vulnerable public web relays. Copy this command and send it to your teammate through a secure messaging channel.

When your teammate executes the command, the invite binds strictly to their Tailscale identity upon first use. This guarantees proper cryptographic attribution so that every prompt entered into the transcript clearly displays who asked for it.

Step 4: Collaborative Prompting and Live Transcript Monitoring

With the connection established, both you and your teammate can view the live terminal transcript in real time as new native turns are persisted. When your teammate types a prompt into their terminal, it routes through the tailnet connection and lands directly in the active agent session as their distinct username.

The system maintains an append-only prompt audit trail, ensuring transparent tracking of all inputs. You maintain full administrative authority over the session, and any requested tool approvals or denials stay strictly under owner control.

mpai: Pros & Cons

mpai Pricing: Free vs Paid

mpai is completely open source and free to use under the MIT license. There are no hidden subscription tiers, cloud workspace fees, or premium feature paywalls associated with the public alpha release. Because it operates locally over your own device network rather than relying on a centralized SaaS platform, you incur no software licensing costs beyond whatever infrastructure you use for Tailscale.

The free open-source distribution includes full access to native session discovery, live terminal transcript streaming, named teammate prompts, role management, and the append-only audit trail. Commercial use is explicitly permitted under the MIT terms, making it accessible for professional engineering teams looking to pilot terminal-based collaboration.

👉 Check the latest pricing and updates on the official website or GitHub repository.

Who is mpai Best For?

For paired cofounders: It provides an effortless way to jump into each other's complex debugging sessions without wasting time copying transcripts or explaining background context.

For remote engineering pairs: It enables real-time terminal collaboration inside familiar CLI tools like Claude Code without forcing teams to adopt a new IDE or browser-based interface.

For security-conscious developers: It leverages private Tailscale tailnets and explicit owner-controlled sharing to keep code discussions off public relays.

Who Should Not Use mpai?

mpai may not be the right choice for engineering teams that rely exclusively on Windows or Linux workstations, as the current public alpha is restricted to macOS hosts. Additionally, developers who do not utilize Tailscale or prefer traditional web-based collaborative coding environments will find that setting up a private device network introduces unnecessary friction.

Teams working in highly regulated enterprise environments that require formal organizational compliance certifications should also exercise caution. Because mpai is currently in a public alpha stage and not yet certified for large organizations, initial rollouts should be kept to non-sensitive pilot projects.

Alternatives to mpai

Traditional screen sharing tools like Zoom or Tuple allow developers to view terminal sessions remotely but lack native multi-user agent attribution.

Cloud-based collaborative coding workspaces offer shared terminal access through web relays but require migrating away from native local CLI agents.

Manual transcript sharing methods let teammates paste conversation histories into messaging apps but break real-time interactivity.

Despite these options, mpai remains the superior choice for developers who want multiplayer terminal AI sessions while staying inside native Claude Code or Codex interfaces without cloud relays.

How We Evaluated mpai

This tutorial and evaluation are based strictly on the official product documentation, public launch materials, security disclosures, and feature statements provided by the maintainers. Our analysis focuses on objective architectural capabilities, installation requirements, and documented user workflows available during the public alpha release.

Final Verdict: Is mpai Worth It?

mpai solves a very specific workflow bottleneck for terminal-heavy developers by bridging the gap between isolated AI agent sessions and team collaboration. Its open-source nature and reliance on Tailscale make it a secure, pragmatic addition for engineering pairs.