Skip to main content
Jentrix/Docs
Open app

Getting started

Go from "I have a Jentrix account" to "my coding agent's work is recorded on a board" in about ten minutes. You do not need a checkout of Jentrix itself, a database, or any server of your own — you are using a hosted deployment, and everything below runs on your machine and talks to it.

Contributing to Jentrix itself, or running your own server? That is a different path — see the repository README for local development, and Deployment for standing up your own instance.

1. Get an account

Sign-up is invite-only. A new account is created only for an email that already has one, that has a pending workspace invitation, or that an operator has allowlisted. There are two ways in:

  • You were invited. Accepting the invitation link creates your account and drops you straight into that workspace.
  • You weren't. Open the deployment and use Request access on the sign-in page — it takes your email and a line about your team and use case, and notifies an operator. You'll hear back out of band; the page deliberately doesn't tell you whether an address already exists.

Once you're allowed in, sign in with Continue with Google or with an email magic link. Both land in the same account — the same address through either route is the same user.

On first sign-in Jentrix creates a personal workspace for you. If you were invited to a team workspace you'll have both; the switcher is top-left.

2. Find your way around

A workspace holds boards, and a board holds tasks in columns. Every task gets a permanent key like STA-123. That much you can drive from the UI right now — the user guide is the full tour.

What makes Jentrix worth wiring up to a terminal is the next part: a coding agent that works in your repository, with its session recorded against a task. That needs three things installed locally — the CLI, the session runner, and a plugin for whichever agent you drive.

3. Install the toolchain

Open Setup in the app (workspace navigation → Setup). It renders every command already resolved for the deployment you're signed into, with nothing to fill in, and reports what the server has actually observed after each step. It is the source of truth; this page explains why each step exists.

Setup's first command is a single line:

bash -c "$(curl -fsSL https://stacks-mvp.vercel.app/install.sh)"

Copy it from Setup rather than from here — that copy names your own deployment. Run it as one command (the bash -c "$(…)" form keeps stdin on your terminal so the browser sign-in and prompts work). It:

  1. installs @jentrix/cli and @jentrix/runner globally via npm — both are required, and jentrix session refuses with RUNNER_NOT_INSTALLED without the runner;
  2. registers the Claude Code plugin and, if codex is on your PATH, the Codex plugin and its MCP server;
  3. signs you in through the browser (jentrix login);
  4. offers to make the current folder session-ready.

It is safe to re-run — npm refreshes the packages, the plugin step switches to its update path, and an existing Codex entry is left alone.

Prefer to do it by hand? npm install -g @jentrix/cli @jentrix/runner && jentrix plugin install is the same thing. Node 20+ required. Full detail, including npx and offline installs, is in CLI install & configuration.

Two local checks Setup can't verify for you, because the server never sees your machine:

jentrix --version
jentrix whoami

4. Connect your agent

The plugin and the MCP server are two halves and neither substitutes for the other: the plugin carries the in-session commands and the trusted lifecycle hooks that identify a session; the MCP server is what gives the agent your Jentrix tools.

Claude Code. The installer registers the plugin, which adds /jentrix-connect, /jentrix-align, /jentrix-plan, /jentrix-checkpoint, /jentrix-status, and /jentrix-end to new sessions (a running session picks them up on restart). For the tools, run /mcp in Claude Code, choose jentrix, and authenticate — Claude Code asks once per checkout.

Codex. The installer registers the server and signs it in. If you installed Codex afterwards, or removed the entry, finish it with:

codex mcp login jentrix

Codex's connection is one global entry per machine, so it follows the deployment the installer saw. Open /hooks in a new Codex task and trust the Jentrix hooks so it can read its session identity.

No credential is written anywhere for either. Your agent authenticates with the deployment over OAuth and you choose the scopes — and optionally pin the credential to one workspace — on the consent screen. MCP servers load at session start, so this applies from your next session.

Headless, or a client with no browser? Mint a personal access token at Account → API tokens (it looks like tm_…) and use jentrix align --pat, or configure the client directly. Per-client instructions for claude.ai, Cursor, and custom SDK agents are in Agent setup variants.

5. Record your first session

From your project folder:

jentrix align

It asks a fixed set of questions — workspace, project, work item, owner — and binds this checkout's session to that work. For Claude Code it also writes ./.mcp.json so the checkout's tools point at this deployment. In-session, the same thing is /jentrix-align (Claude Code) or $jentrix-align (Codex).

Then work as you normally would. When you're done:

jentrix push plan < plan.md    # store a typed artifact against the task
jentrix session end            # push final artifacts, verify capture

The session lands on the project's Runs tab with its transcript, artifacts, and cost. Run your first connected session walks the whole loop with the reasoning behind each step.

No repository yet, or want the structure proposed for you? Set up your first project has a copy/paste prompt that analyzes a repo and creates the board, project, and anchor task after you approve.

Where to go next