Skip to content
Desktop guide

How to set up Sydla Desktop with Claude Code or Codex: a step-by-step guide (2026)

A complete setup guide: install the CLI and Sydla Desktop, sign in once, pick the agent, model and access, start a thread from a piece of work with its brief and UX states, check each state in the preview, and review the diff before anything ships.

Sydla Journal cover: set up Sydla Desktop with Claude Code or Codex. Six steps in a Desktop window: install (.app, .deb, .rpm, AppImage or .exe), connect with a browser sign-in shared with the CLI, pick claude or codex already signed in, choose Read only, Edit files or Full access, start from a piece of work with its brief and UX states, and review Changes to commit and open a merge request. $19 per user per month, no AI credits, no token markup.

Your agent is paid for. The context around it is not.

Most product teams in 2026 already pay for a coding agent. Someone has Claude Pro or Max for Claude Code, someone else has ChatGPT Plus for Codex. The agent is good. The problem is everything it has to read before it writes a line: the ticket in Jira, the spec in Confluence, the research in Notion, the screens in Figma, the flow on a Miro board and the launch visual in Canva.

Each of those tools now sells its own AI with its own meter. Rovo credits, Notion credits, Figma AI credits, Miro credits, Canva's allowance. We called it AI credit fatigue, and it has a twin: six MCP servers, six logins and six rate limits, so every ticket costs the agent six context hops before it starts.

This guide shows the other setup. You keep Claude Code or Codex on the plan you already pay for, and you run it from Sydla Desktop, where the brief, the acceptance criteria and the UX states of the work sit in one place. It replaces our short first Desktop guide with the full setup, step by step, as of 20 November 2026.

What the six-tool setup costs before the agent runs

Here is the stack an agent usually has to read, one paid seat per tool, with the built-in AI limit each one adds. Each tool has a full page on our comparison hub.

One paid seat per tool, monthly USD list prices before tax, checked on 14 November 2026. AI limits checked on 18-20 November 2026.
Tool and planSeat / monthBuilt-in AI limitWhat the agent needs from itHandoff
Jira Standard$9.7025 Rovo credits per user, pooled; $0.01 per extra credit from 3 December 2026Ticket and acceptance criteriaRovo MCP, enriched calls metered
Confluence Standard$7.1525 Rovo credits per user, same poolSpec and decisionsSame Rovo MCP, same pool
Notion Plus$12.00Custom Agents need Business or Enterprise, then $10 per 1,000 creditsNotes and researchNotion MCP, separate login
Figma Professional, Full seat$20.003,000 AI credits a month, no rollover; MCP reads 200 a dayScreens and statesFigma MCP, daily cap
Miro Starter$10.0025 credits per member, pooled; AI pauses at zeroFlows and diagramsMiro MCP, 100 calls a day on Starter
Canva Business$25.00Per-member AI allowance, top-ups on BusinessLaunch visualsCanva connector, separate login
Six-tool total$83.85Six meters, six resets, six top-up pagesSix places to keep in syncSix context hops per ticket
[Sydla](/)$19.00None: your Claude Code or Codex plan pays for the model, no token markupOne work item: brief, criteria, states, runsOne Sydla connection
Bar chart of monthly list prices per seat in November 2026: Jira Standard $9.70, Confluence Standard $7.15, Notion Plus $12.00, Figma Professional Full seat $20.00, Miro Starter $10.00 and Canva Business $25.00, a total of $83.85 with six AI meters, against Sydla at $19.00 with no AI credits. The saving is $64.85 per user per month.
Six seats and six meters, or one workspace where your own agent plan pays for the model.

That is $64.85 per user per month less, before you count the credit top-ups. The pricing section lists every Sydla plan, and the comparison pages are honest about where each specialist tool still wins. The setup below is where the second saving comes from: the agent reads one work item instead of six tools.

Before you start: what you need

The setup takes about fifteen minutes if your coding agent is already installed. You need four things:

  1. A coding agent CLI, signed in. For Claude Code, run curl -fsSL https://claude.ai/install.sh | bash on macOS or Linux, or irm https://claude.ai/install.ps1 | iex in Windows PowerShell; Homebrew, WinGet and npm work too. Run claude once and sign in. Claude Code needs a Pro, Max, Ultra, Team, Enterprise or Console account, not the free plan. For Codex, run curl -fsSL https://chatgpt.com/codex/install.sh | sh, then codex, and choose Sign in with ChatGPT. Codex is included in ChatGPT Plus at $20.
  2. Git and a repository. Desktop works in Git folders, and each thread's changes are committed per repository.
  3. A Sydla account, optional at first. Desktop runs local threads without one. You need the account for initiatives, shared briefs, UX Studio and team plugins. Create it here.
  4. glab, optional. If your code lives on GitLab.com, glab auth login lets the agent show merge request and pipeline cards, and lets Desktop open merge requests for you.

Step-by-step: from install to first reviewed change

Here is the whole path, in the order you will do it the first time.

  1. Install Sydla Desktop. Sign in to Sydla and open Install in the dashboard. It offers the build for your system: .exe or .msi on Windows, an .app archive on macOS, and .deb, .rpm or AppImage on Linux. Chromium is bundled, so you need no separate browser. Installed apps check for updates every six hours and never restart while an agent works.
  2. Connect your account. Open Desktop and choose Connect to Sydla. Sign-in happens in your browser, and the key is stored in the same file as the Sydla CLI, so one sign-in covers both. Choose Continue without an account if you only want local threads for now.
  3. Pick the agent, model and effort. Desktop finds the claude and codex commands on your machine and loads their model lists from your own sign-in and configuration. The model popup lets you pick the provider, the model and the reasoning effort. If a list looks stale, Settings, Models & keys, Refresh reloads it. You can also save a persona: a name, a role, the agent and model behind it, and instructions sent before every turn.
  4. Add your work. With an account, a thread starts across your Sydla workspace. The agent lists the repositories it can reach and prepares only the ones it needs: an existing folder is reused, a missing one is cloned with a short-lived token that is not kept. A folder whose Git remote belongs to a Sydla project shows that project's key.
  5. Choose the access. Each thread has Read only, Edit files or Full access. Start with Read only to frame a change. At Edit files, Claude Code asks before most shell commands, and the thread shows a card to allow once, always allow that tool in this thread, or refuse. Codex has no such hook, so at Edit files it stays inside its own sandbox.
  6. Start from a piece of work. On the Initiatives page, open the plan and start a thread from the piece you want built. The thread starts in the linked folder, the agent gets Sydla's tools for each turn (a private file for Claude Code, configuration overrides for Codex), and its first message names the initiative context to read: the brief, the criteria and the published UX handoff.
  7. Check each state. Ask the agent to start the dev server. The Preview panel offers the local addresses it printed in one click, and the agent's own browser can open each state and compare it with the design. Take over pauses the agent's browser if you need to sign in yourself, and Give back lets it continue on the same page.
  8. Review Changes, then ship. The thread shows every message, tool call, input and output, and how each turn ended, with duration, tokens and cost. Open Changes to read the diff of the selected repository, then commit and push. On GitLab.com, the Finish & Ship quick action commits, pushes and opens a merge request, and the merge and the tag each need their own confirmation.
Four steps from brief to review in Sydla Desktop: a shared brief with numbered acceptance criteria in Sydla Docs; empty, loading, error and success states in UX Studio; a Desktop thread running your local Claude Code or Codex CLI with Sydla's tools for each turn, in Read only, Edit files or Full access, with Preview and its own browser; and Changes, where a person reviews the diff against the criteria and states, then commits, pushes and opens a merge request.
The model runs on your plan. The context comes from one work item.

A worked example: the billing export

Take a real ticket: admins need to export invoices as CSV. In Sydla Docs the brief has the goal, the users, what is out of scope, and numbered criteria. AC1, the columns come in a fixed order. AC2, an empty date range shows the empty state with a hint, not a blank file. AC3, commas, quotes and accented names survive the export. Our acceptance criteria guide explains why each one names a check you can run.

In UX Studio, the designer draws four states next to the happy path: empty, loading, error and success, then publishes the handoff on the same initiative. Our design-to-code handoff guide covers why states beat screenshots for an agent.

In Desktop, you start a thread from that piece of work with Codex at Edit files. The agent reads the brief and the handoff through Sydla's tools, writes the export, runs the tests and opens each state in its browser. In our recorded run, 11 tests covered column order, empty input, missing values, CSV escaping and Unicode. You read the diff in Changes next to AC1 to AC3, ask for one follow-up in the same thread, then commit. The agent's last message starts the review. It does not end it.

Designers can join in Desktop too. Opening a screen in Sydla, UX shows the real canvas inside the app with the agent panel beside it. In Edit live mode the agent's changes land on the shared design at once, with its cursor labelled by your initials and the persona's name. Suggest and Comment only refuse those writes.

Prefer to wire the agent by hand?

Desktop connects Sydla's tools for you on every turn. If you also run the CLI outside Desktop, in a terminal or an editor such as Cursor or VS Code, add the same Sydla MCP server yourself from the MCP page of your dashboard. In Claude Code, use claude mcp add --transport http sydla https://sydla.com/mcp with an Authorization header carrying your key, and pick the local, project or user scope. In Codex, add an [mcp_servers.sydla] table to config.toml with the same url and the key in http_headers.

It is one server for the brief, the plan, the criteria and the UX handoff. Compare that with the six servers in our MCP servers guide, each with its own login and its own limits. Whichever way you connect, treat what the agent reads in tickets and pages as data, not instructions; our prompt injection guide lists the guardrails.

When something does not work

Most first-run problems come from the CLI, not from Desktop:

  1. The provider is greyed out. Desktop could not find the command. Check claude --version or codex --version in a terminal, then refresh the model list.
  2. The thread reports an expired sign-in. Run claude or codex in a terminal once and sign in again. Desktop shows the failure in the thread instead of hiding it.
  3. The agent stops for every command. That is Claude Code at Edit files asking for approval. Always allow the tools you trust for that thread, or raise the access once the plan is clear.
  4. Scheduled runs did not happen. Schedules run in the background while Desktop is open in the tray. After a restart, open Desktop; missed repetitions become one run.
  5. Launched work never starts. Running work launched in Sydla is off by default. Turn it on in the Initiatives page; each run then stays under the launch's cost cap.

What it costs, all in

Two lines, both flat. Sydla is $19 per user per month for the workspace: briefs, plan, UX Studio, Desktop threads and review. Your agent plan is separate and stays yours: Claude Pro at $20 a month, or $17 billed yearly, Max at $100 and Ultra at $200, or ChatGPT Plus at $20 for Codex. Those plans have usage limits, but no meter on every token, and Sydla adds no markup on top. Our cost per ticket guide works through what a ticket really costs on each.

If you are still deciding between the two agents, Claude Code vs Codex for product teams compares them on the same brief. If you are a team of one, the solo founder stack guide does the yearly maths. And if you are leaving the six tools, the 30-day migration plan says what to move first.

Set it up on the next ticket

Pick the next feature on your list. Write its brief and criteria in Sydla, draw its states, install Desktop, and start a thread from that piece of work with the agent you already pay for. The first run tells you more than any comparison table.

Start your $19/mo AI SDLC workspace and keep your Claude Code and Codex accounts exactly as they are.

Frequently asked questions

What is Sydla Desktop?

Sydla Desktop is a desktop app for Windows, macOS and Linux that runs the coding agent installed on your machine, Claude Code or Codex, in threads you can follow live. Each thread shows every tool call, lets you stop or follow up, and ends in a diff you review. Connected to a Sydla account, a thread can start from a piece of work and read its brief, acceptance criteria and UX states.

Do I need a Sydla account to use Sydla Desktop?

No. Desktop runs local threads with your existing Claude Code or Codex access without a Sydla account; provider charges still apply. You need an account, $19 per user per month, for initiatives, shared briefs, UX Studio designs, team plugins and starting threads from planned work.

Does Sydla charge for Claude Code or Codex tokens?

No. Sydla Desktop runs the claude or codex command you are already signed in to, so the model usage counts against your own Claude or ChatGPT plan. Sydla sells no AI credits and adds no token markup.

How do I connect Claude Code to Sydla?

Install Claude Code with curl -fsSL https://claude.ai/install.sh | bash, run claude and sign in with a Pro, Max, Ultra, Team, Enterprise or Console account. Then install Sydla Desktop and pick Claude Code in a new thread; Desktop gives it Sydla's tools on each turn. Outside Desktop, add https://sydla.com/mcp with claude mcp add --transport http and your key in an Authorization header.

How do I connect Codex to Sydla?

Install Codex with curl -fsSL https://chatgpt.com/codex/install.sh | sh, run codex and choose Sign in with ChatGPT. Then pick Codex in a Sydla Desktop thread; Desktop passes Sydla's tools as configuration overrides. Outside Desktop, add an mcp_servers.sydla table to config.toml with the Sydla MCP url and your key in http_headers.

Real native session with Codex. Edited waiting times; 11 acceptance tests pass. No commit or push is performed.

Keep reading.

All articles

ONE WORKSPACE. THE ENTIRE SDLC.

Your next project.
One $19 subscription.

Keep using your Claude Code and Codex accounts. Bring the work into Sydla.

Get SydlaUSD / user / month · Billed monthlyProvider subscriptions remain separate.