A practical guide to running Claude Code across a whole team, covering the shared setup that makes it work, the skills our CX and ops teams run every day, and how to get started this week.
Most teams treat AI as a personal productivity tool. The bigger gains show up when a whole team works from the same setup.
Claude Code for teams is a shared setup where everyone works from the same context layer, connects to the same real tools and data, and authenticates securely. In our live session, that setup cut a one-hour task to 5 minutes and spread useful automations across a team in days. So, what does this set up look like?
A shared context layer (a GitHub repo, or a scrappy shared folder) lets a whole team run the same skills and tools.
Connecting Claude Code to real tools like Slack, Notion, Email, and BigQuery is where team productivity grows.
Secure access uses controls you already have: secrets in a password manager, logged queries, and daily multi-factor authentication.
Lorikeet's implementation team builds kickoff docs in 5 minutes instead of an hour, using one shared skill.
You do not need a top-down mandate. One person with a shared folder and a single useful skill is enough to start.
In our first Claude Code session, we covered what the tool is like for one person. This follow-up is about the harder, more valuable version, using it across a team. Lorikeet runs as an AI-native company, and the day-to-day value shows up when everyone shares skills, context, and data access. Robbie Tilleard, GM EMEA at Lorikeet, and Estelle Berton, a Forward Deployed PM, walked through how we set this up, the automations our technical and nontechnical teams rely on, and the mistakes to avoid.
What does it take to use Claude Code as a team?
Using Claude Code as a team takes three things: a shared context layer everyone works from, real access to your tools and data, and secure authentication. Doing this solo is easy. Doing it as a team takes a little setup, and that is where the productivity compounds.
This applies to Claude Code, Codex, or any agent tool your team uses. Solo adoption is low-stakes, you buy a subscription and try it on a weekend with no coordination. The team version needs structure so everyone runs the same skills and follows the same practices, even while working independently.
How do you build a shared context layer?
A shared context layer is one place your whole team taps into for context, skills, and tools. The ideal version is a single GitHub repo, because it adds version control. A scrappy version, a shared Drive, Dropbox, or Notion folder, captures most of the value without it.
The ideal setup: one GitHub repo
At Lorikeet, our CEO set up a shared repo. It holds the team's context, a folder of skills, a team directory, and related data files. A CLAUDE.md file describes the repo as the team's shared layer: clone it, pull it every morning, and never hard-code personal logins or customer data.
The scrappy setup: a shared folder
You do not need GitHub to start. In the session, Robbie built a skill on his machine, then pushed it to a plain folder on his desktop. The team loses version control, but the skill still spreads. The point is a shared space where everyone runs the same skills.
"We should start the day with agents telling us what to do, and scale that across the team so we are all working from the same source of truth." — Estelle, Lorikeet
How do you connect Claude Code to your real tools and data?
You connect Claude Code to real tools through MCP integrations or direct APIs. Model Context Protocol (MCP): a standard way to connect an AI tool to an app like Slack or Notion so it can read and act on real data. Tools like Slack, Notion, Linear, and Grain have native MCP connections you can set up without engineering help. Connecting to ticketing systems, your CRM, or BigQuery is straightforward; reaching private production data in Azure or Google Cloud needs engineering support.
This is where team productivity grows, because the work gets tied to your actual data. Estelle built proactive monitoring on Lorikeet's BigQuery data: a CLAUDE.md file that tells Claude to send a Slack message when a key metric moves. Start with native MCP connections, then add the harder integrations as you get engineering time.
How do you keep it secure?
Securing team access to Claude Code is doable with controls your organization already uses. At Lorikeet, accessing customer data requires a legitimate business reason, every query is logged, and access uses daily multi-factor authentication. Secrets live in a password manager, not in the shared repo.
Treat the shared repo as visible to your whole team, holding context, skills, and tools, with no secrets and no real customer data inside it. Authentication and compliance are manageable, and most of the controls are ones your security team already runs.
What can a team actually build with Claude Code?
Teams build skills that run the same way for everyone. At Lorikeet, common ones include a daily Slack digest, a weekly priorities skill, BigQuery dashboards, and a sales-handover-to-kickoff workflow that cut a one-hour task to 5 minutes. Each skill lives in the shared repo so the whole team can run and improve it.
Daily digest. A skill reads the Slack channels that matter and posts a 24-hour summary to the team channel, so people stop getting lost across dozens of channels. It runs automatically as a scheduled routine in the Claude desktop app.
Weekly priorities. A skill reads production metrics, Linear tickets, Slack, and Grain recordings, applies a stack rank, and posts the week's priorities. The team reviews it every Monday and aligns before executing.
Dashboards on demand. A CLAUDE.md file tells Claude how to read BigQuery data and chart it. Any leader can run the report and get the same dashboard, work that used to need engineering time.
Sales handover to kickoff. One skill turns meeting recordings, Slack threads, and emails into a handover doc. A second turns that handover into a kickoff doc. Together they took a task from about an hour to 5 minutes.
How to design a skill
The fastest way to build a skill is to make the output first. Create the Notion or Google doc you want, point Claude Code at it, and ask it to build a skill that produces that output from a set of inputs. As Estelle described it, no one writes skills line by line anymore, because that is what Claude is good at.
These skills took our team from hours to minutes on repeat work. See how Lorikeet builds AI that works across your whole support operation.
Lorikeet's take on adopting Claude Code as a team
At Lorikeet, we have found that team adoption does not need a top-down mandate. The pattern that works is bottom-up. In the session, Robbie told the story of a salesperson who on a Tuesday asked why he would use Claude Code over ChatGPT, and by Thursday had automated his outbound motion and built skills for his whole sales team.
"You don't need a CEO to drive this. Someone has to own it, and if you're here, maybe it's you." — Robbie, Lorikeet
Keeping a human in the loop is the other piece. As Estelle put it, the further away that humans get from agents right now, the worse the output. Agents drift, so verify a new skill when you build it and re-check on a regular cadence. Before pushing a skill to the shared repo, ask Claude what it changes. Whenever Robbie ships one, he asks whether there is any blast radius he should know about. The same shared-skill approach sits behind how Lorikeet scales support.
Lorikeet is the AI Concierge for complex, regulated support, resolving multi-step tickets end-to-end across chat, email, and voice inside your existing systems.
Key takeaways
Team adoption of Claude Code rests on three things: a shared context layer, real tool and data access, and secure authentication.
A GitHub repo is the ideal shared layer for version control; a shared Drive or Notion folder works to start.
Start with native MCP tools (Slack, Notion, Linear, Grain) that need no engineering help, then add harder data sources.
Verify agent output when you build a skill and re-check regularly. Agents drift, and human judgment still matters.
You can start bottom-up this week: one shared folder, a few skills, one team.
Frequently asked questions
How much does Claude Code cost for a team?
Claude Code is part of your Anthropic subscription, so team cost depends on your plan and usage. The team setup itself, a shared repo plus skills, adds no extra license cost. You can even start with a shared Drive or Notion folder and no new seats.
How long does it take to set up Claude Code for a team?
A scrappy setup takes about a day: create a shared folder, add a few skills, and have the team run them. A full version-controlled repo with real data access takes a technical person a couple of days, plus engineering time for private production data.
Can non-technical teams use Claude Code?
Yes. Estelle opened a terminal for the first time in early 2026 and now builds and ships team skills. Starting with native MCP tools like Slack, Notion, and Linear means non-technical operators can build useful automations without engineering support.
What's the difference between using Claude Code solo and as a team?
Solo use is one person running the tool on their own machine. Team use adds a shared context layer so everyone runs the same skills, connects to the same tools, and follows the same practices, which is where the productivity compounds across the group.
Do you need engineering help to connect Claude Code to your data?
Not to start. Tools with native MCP connections (Slack, Notion, Linear, Grain) connect without engineering. You need engineering support when you reach private production data in places like Azure or Google Cloud, or when setting up authentication for sensitive systems.
How do you stop a shared skill library from getting messy?
Use version control and verification. Ask Claude what a skill changes before pushing it, check for blast radius, and build skills that prompt the team to keep shared context current. Lorikeet uses a weekly-updates skill that asks whether customer files need cleaning up.
Is Claude Code worth it for CX and ops teams?
For teams doing repeat work like reporting, handovers, and monitoring, the time savings add up fast. In our session, a sales-handover-to-kickoff workflow cut a task from about an hour to 5 minutes, and the same skill runs for everyone on the team.
Where to start
Using Claude Code as a team is less about the tool and more about the setup around it. A shared place to work from, real tool access, and sane security are what let a whole team move at the speed one person can. The early movers are still few. In the session, Robbie put teams using these tools this way in the top 1% of adoption worldwide, and said that gap will compound fast over the next five years.
You do not need a perfect setup or a top-down program. Create a shared folder, add a couple of skills, connect a native MCP tool, and get one team running this week. Then verify the output, keep a human in the loop, and grow from there.
Want to see what AI built this way looks like in production? Get a demo of Lorikeet, or read the first session recap to start from the basics.
Book a call
See what Lorikeet is capable of
