How to Use Claude Code With Your Team

Claude Code started as a tool one developer runs in one terminal. Most teams now have several people running it at once, and some also run OpenAI’s Codex. That raises practical questions. How do you make every agent follow the same rules? How do you share tools without sharing secrets? How does anyone see what the others are doing?

This guide covers what Anthropic’s official documentation confirms about running Claude Code as a team, how to fit Codex into the same setup, and where the gaps are.

Which plans include Claude Code for teams

Claude Code requires a Pro, Max, Team, Enterprise or Console account. The free claude.ai plan does not include it, according to Anthropic’s setup guide.

For organizations, the authentication docs list two main routes:

• Claude for Teams or Enterprise. Anthropic recommends this for most teams. Members get Claude Code and Claude on the web with centralized billing and team management. Teams is self-service and includes admin tools, SSO and server-managed settings. Enterprise adds domain capture, role-based permissions and the compliance API.
• Claude Console. For organizations that prefer API-based billing. Admins invite users and assign a role, such as a Claude Code role that can only create Claude Code API keys.

Cloud providers such as Amazon Bedrock and Microsoft Foundry are also supported. For current seat prices, Anthropic points to claude.com/pricing.

Put shared instructions in a CLAUDE.md

The single most useful team habit is a project CLAUDE.md. Claude reads it at the start of every session. Anthropic’s memory docs describe the project file, at ./CLAUDE.md or ./.claude/CLAUDE.md, as team-shared instructions that reach teammates through source control.

Good content for it: build and test commands, coding standards, architectural decisions, naming conventions and common workflows. Personal preferences belong elsewhere. Each person can keep a CLAUDE.local.md that you add to .gitignore, or a user-level file at ~/.claude/CLAUDE.md.

Two tips from the docs. Run /init to generate a starting file from your codebase. And add to the file when Claude makes the same mistake twice, or when a new teammate would need the same context to be productive.

Larger organizations can also deploy a managed CLAUDE.md to a system location on every machine through MDM or similar tools.

Share settings and permissions in .claude/settings.json

Claude Code reads settings from several files, each with its own scope. Per the settings docs:

• .claude/settings.json is the shared project file. Commit it so teammates get the same permissions, hooks, plugins and environment variables.
• .claude/settings.local.json holds personal overrides for one project and stays out of your commits.
• Managed settings come from your organization and sit above everything else.

Permission rules come in three kinds: allow, ask and deny. The permissions docs say rules are evaluated deny first, then ask, then allow, and that a broad deny cannot be overridden by a narrower allow. One point worth repeating to your team: permission rules are enforced by Claude Code, not the model. Writing “never run X” in CLAUDE.md shapes behavior, but a deny rule is what actually blocks it.

Share tools with a project .mcp.json

MCP servers connect Claude Code to outside tools and data. Project-scoped servers live in a .mcp.json file at the repo root. The MCP docs say to check it into version control so everyone gets the same MCP tools. Add one with claude mcp add --scope project.

The file supports environment variable expansion, such as ${API_KEY}, so the shared config can reference a key without containing it. Each teammate sets their own value. For security, Claude Code asks for approval in interactive sessions before it uses project servers from .mcp.json.

Share skills, slash commands and subagents

If your team keeps pasting the same checklist into chat, make it a skill. Skills are SKILL.md files that Claude uses when relevant, or that you call directly as /skill-name. According to the skills docs, custom commands have been merged into skills, and existing files in .claude/commands/ keep working. Save a skill in .claude/skills/ and commit it, and your team gets it too.

Subagents work the same way. A subagent is a specialized assistant with its own context window, system prompt and tool access. Project subagents in .claude/agents/ can be checked into version control so the team can use and improve them together.

To bundle skills, subagents, hooks and MCP servers into one installable unit, Claude Code has plugins, and a team can run its own plugin marketplace.

Enforce rules with hooks

Hooks are shell commands that Claude Code runs at set points in its lifecycle. The docs frame them as deterministic control: the action always happens, instead of relying on the model to remember. Examples from the guide include running Prettier after every file edit and blocking edits to protected files. Put team hooks in the shared .claude/settings.json so they run for everyone.

Run Claude Code in GitHub Actions and CI

The Claude Code GitHub Action lets anyone on the team mention @claude in a pull request or issue comment to have Claude analyze code, make changes and push commits. Run /install-github-app from Claude Code for the quick setup. The action follows your repo’s CLAUDE.md, so the rules you wrote once apply in CI too.

For other pipelines, the -p flag runs Claude Code non-interactively, as described in the headless docs. Team and Enterprise plans can also turn on Code Review, which posts inline findings on pull requests without approving or blocking them.

Admin controls, analytics and spend

For admins, Anthropic documents several team features:

• Analytics. Teams and Enterprise get a dashboard with usage metrics, contribution metrics (public beta) through a GitHub integration, a leaderboard and CSV export. Console customers get usage, spend and team insights. See the analytics docs.
• Spend controls. Teams and Enterprise admins set spend limits in admin settings; Console uses workspace spend limits, per the costs docs.
• Telemetry. Claude Code can export usage, cost and tool activity through OpenTelemetry.
• Onboarding. The /team-onboarding command turns your recent usage into a guide a teammate can paste as a first message, according to the commands reference.

The gap: every agent lives in its own terminal

Everything above makes each person’s agent behave the same way. None of it puts those agents in one room. Each person’s local Claude Code runs on their own machine, with its own context. A teammate cannot ask your local agent a question, and two people’s local agents cannot hand work to each other in a thread everyone can read.

Anthropic does offer shared surfaces, and they are worth knowing. Claude Tag, in public beta on Team and Enterprise, lets anyone in a Slack channel tag @Claude with a task. It runs in an ephemeral sandbox, not on your computer. Agent teams, an experimental feature, let one Claude Code session lead several teammate instances, and cross-session messaging passes notes between your own sessions. All useful. Neither is a place where each person’s own local agent, Claude Code or Codex, sits next to the people it works with.

Bringing your team’s agents into one chat

Conduyt Team Chat sits inside the Conduyt CRM: channels and DMs next to your deals, contacts and files. The AI teammates part is rolling out now, so not every account has it today.

Here is how it works:

• Pair your own agent. Each person can pair their own Claude Code, Codex or Grok Bot agent as an “AI teammate” with its own name and @handle. Pairing uses a one-time code pasted into a small listener on that person’s computer. The code is single use and expires after 15 minutes. No API key is needed.
• It runs on your machine. The agent runs on its owner’s computer, using the owner’s own AI subscription, with nothing extra from Conduyt.
• Anyone can ask it. Teammates can @mention any person’s agent from anywhere, and it answers in the thread with live CRM data.
• Same permissions as its owner. An AI teammate acts with its owner’s Conduyt permissions, never more.
• A person approves outgoing messages. Anything an AI wants to send under a person’s name needs that person’s approval: Approve, Edit first or Decline.
• Agents can work together, with a brake. Agents can @mention each other. After six AI turns in a row, the thread pauses until a person weighs in.

Team Chat is on every plan with unlimited users. AI teammates are on the Professional plan ($499/mo billed monthly, or $399/mo billed annually) and in the 20-day free trial, with no credit card, as the rollout reaches your account. If you are new to the idea of agents acting inside a CRM, read what agentic CRM means or Bring Your Own AI.

Using Codex and Claude Code together on one team

Many teams run both, by preference or with one writing and the other reviewing. The goal is that both agents follow the same rules and reach the same tools.

Share one instructions file. Codex reads AGENTS.md before it starts work. Per OpenAI’s AGENTS.md guide, it checks a global file in ~/.codex, then walks from the repo root to your current directory, and files closer to your directory take precedence. Claude Code reads AGENTS.md directly when the repo has no CLAUDE.md (version 2.1.277 or later), per the memory docs. If you also want Claude-specific rules, create a CLAUDE.md whose first line is @AGENTS.md and add the extra rules below it. Both agents then read the same core instructions.

Give both agents the same tools. Both tools speak MCP. Claude Code uses the project .mcp.json, and the Codex CLI adds local or remote MCP servers with codex mcp. Any server your team depends on, such as a CRM’s MCP server, can be added to both.

Keep the hand-offs visible. When one agent writes and another reviews, the review is only useful if the people involved can see it. That is the case for putting both agents in a shared thread rather than copying output between terminals.

A quick team setup checklist

1. Pick a plan route: Claude for Teams or Enterprise, or Console.
2. Commit a project CLAUDE.md, or an AGENTS.md if you also run Codex.
3. Commit .claude/settings.json with your allow, ask and deny rules and team hooks.
4. Commit .mcp.json with environment variables in place of secrets.
5. Move repeated procedures into .claude/skills/ and .claude/agents/.
6. Turn on analytics and spend limits.
7. Give people and agents one shared place to talk.

Frequently asked questions

Can several people use Claude Code on the same project?

Yes. Each person runs Claude Code on their own machine, and the team shares behavior through files in the repo: a project CLAUDE.md, .claude/settings.json, .mcp.json, and shared skills and subagents. Personal overrides go in CLAUDE.local.md and .claude/settings.local.json.

How do I share Claude Code settings with my team?

Commit .claude/settings.json at the project root. Anthropic’s docs say everyone who clones the repo then gets the same permissions, hooks and plugin settings once they trust the folder. Organization-wide policy goes in managed settings, which user and project settings cannot override.

Which Claude plan should a team use for Claude Code?

Anthropic recommends Claude for Teams or Enterprise for most teams. Both include Claude Code with centralized billing and team management. Organizations that prefer API-based billing can use the Claude Console instead. Check claude.com/pricing for current prices.

Can Codex and Claude Code use the same instructions file?

Yes. Codex reads AGENTS.md. Claude Code reads AGENTS.md when there is no CLAUDE.md, or you can import it by starting your CLAUDE.md with @AGENTS.md.

Can my teammates’ Claude Code agents talk to each other?

Claude Code’s built-in coordination centers on one person’s sessions. Cross-session messaging lets Claude message your own other sessions, and the experimental agent teams feature has one session lead several teammates. In Conduyt Team Chat, which is rolling out AI teammates now, paired agents can @mention each other in a thread, and the thread pauses after six AI turns in a row until a person weighs in.

Do I need an API key to add Claude Code to Conduyt Team Chat?

No. You paste a one-time pairing code into a small listener on your computer. The code is single use and expires after 15 minutes. The agent runs on your computer with your own AI subscription, and nothing extra from Conduyt.