VibeFlock in two minutes: connecting Claude, Claude Code and Codex
By Soumen Sarkar · October 2026
Playbook. Connect Claude, Claude Code and Codex to one team record in two minutes, so every session starts knowing what the team is doing.
How do you connect Claude, Claude Code and Codex to one team?
You connect every agent to the same shared record, once per person, and from then on each session starts knowing what the team is doing. With Campfyrr that takes a couple of minutes: Claude adds it as a custom connector and signs in, Claude Code runs one pasted line that adds both the connection and the hooks, and Codex takes a short block in its config file. Everything you need is on the connect page.
This is the first step of the first principle, keep moving. An agent can only keep going without stopping to ask if the context reaches it on its own. Connecting is how the context starts to travel.
Who is this for, and what do you get?
It's for a tech lead setting up a team where every person runs several agents, and for any developer who wants their sessions to stop starting from nothing. You can do it alone in your own repository first, then bring the team in.
When you're done:
- Every Claude Code session on your machine starts with a brief of what changed since your last session.
- The first time a session reads or edits a part of the code, it's told what the team knows about that part.
- Claude in the app or on the web, Codex and Cursor can all call the same tools to search the team's record, state assumptions and leave handoffs.
- Your teammates see what your sessions are working on, without you writing a status update.
What should the team agree before anyone connects?
A few choices make the set-up land well, whatever tools you use. They take ten minutes of conversation and save a lot of confusion later.
- One record for the whole team. Decisions, rules, assumptions and handoffs go to one place every agent can search. If they're spread across chat, pull requests and people's heads, no agent finds them.
- One project per repository. The record is easier to search when each codebase has its own project. Campfyrr's Claude Code hooks name the project after the folder the session runs in, so keep the repository's folder name the same on every machine.
- One set of standing instructions. Put the same short "how we work" section in CLAUDE.md or AGENTS.md, so every agent behaves the same way from its first turn. The team agent instructions template is a starting point.
- Who connects what. Most people will connect Claude Code on their own machine. Anyone who uses Claude in the app or on the web adds the connector. Anyone on Codex or Cursor connects with a token.
How do you connect Claude in the app, on the web, or in Claude Code in the cloud?
Claude connects without a token. It signs you in and asks which team it may act in. Do it once, and it works in every Claude session: the app, the web, and Claude Code in the cloud.
- On the connect page, under Add to Claude, copy Campfyrr's address.
- Open Claude's connectors (the connect page links straight to them), choose Add custom connector, name it Campfyrr and paste the address.
- Click Connect, sign in, pick your team and click Allow.
- Start a new session. Campfyrr's tools are there from then on.
The connector gives Claude the tools. The automatic hooks, which brief a session without anyone asking, come from the Claude Code set-up below.
How do you connect Claude Code on your own computer?
This is the set-up most people want, because it adds the hooks as well as the tools.
- Sign in and pick the team. On the connect page, check that the right team is selected in the header. The page tells you whether you're about to get a team token (your agent acts as you on your team) or a personal token (your agent only sees what it creates). For team work you want the team token.
- Generate the token. Under "Or connect with a token", click Generate connection token. The token is shown once, so copy what you need now.
- Choose Claude Code and "One paste (recommended)". Leave Post a work log to Team updates ticked if you want each session to file a short work log; untick it if you don't.
- Paste the line into any terminal. PowerShell, cmd, bash and zsh all work. It needs Node 18 or later, which Claude Code already uses.
- Read what it prints. It finishes by showing the top of the brief your next session will see.
- Start a new Claude Code session. Sessions that were already open don't pick up the new hooks.
What the paste does, in order:
- It checks the token with the server before it changes anything, and says whether it's a team token or a personal one.
- It saves the token and the app's address in a config file in your home folder that only you can read.
- It downloads the session-brief hook, the work hook and, if you kept it, the work-log hook.
- It adds those hooks to your Claude Code settings file, keeping everything else there and saving a backup of the old file first. Running it twice still leaves one of each.
- It registers Campfyrr's tools for all your projects.
- It runs the brief once, as Claude Code will, and prints it.
What about Claude Code sessions that start from a fresh checkout?
Claude Code on the web, a CI runner or a new machine starts from a clean copy of the repository, without anything in your home folder. For those, choose In a repo (cloud) on the connect page and run the line in the repository's folder.
It writes the connection, the hooks and a short note telling agents to keep the record (in CLAUDE.md) into the repository itself. None of those files holds the token, so you can commit them. Then set the CAMPFYRR_TOKEN environment variable wherever the sessions run; for Claude Code on the web, that's the cloud environment's settings. Each person should use their own token there, so the work is filed as them.
How do you connect Codex?
Generate a token as above, then choose Codex CLI. The page gives you a block to add to ~/.codex/config.toml. It has this shape, with your own address and token filled in:
[mcp_servers.campfyrr]
url = "<the address the connect page shows>"
bearer_token = "<your token>"
Restart Codex after adding it. Field names for remote connections have shifted between Codex versions, so if it doesn't take, run codex --help or check its docs. Codex doesn't get the automatic hooks, so ask it to call session_brief at the start of each session, or put that in AGENTS.md.
How do you connect Cursor and other MCP clients?
Choose Other (JSON). You get an mcpServers block with Campfyrr's address and an Authorization header carrying your token. Most MCP clients, including Cursor, Windsurf and VS Code, accept that shape in their own MCP config file; check your client's docs for where the file lives and the exact key names. Like Codex, these clients call the same tools but don't get hooks, so the standing instructions do the work.
If you only want the tools in Claude Code, without hooks, MCP only gives you a single claude mcp add line. Paste it, restart Claude Code, and your agent gets the brief when it calls session_brief, not automatically.
What instructions should every agent get?
Hooks brief Claude Code automatically. Every other client relies on what you tell it, and Claude Code benefits from the same reminders. Put a short section like this in CLAUDE.md or AGENTS.md:
## Team record
At the start of a session, call session_brief for this project.
Before asking anyone, search the team's record (search_memory).
When something is unclear and the record doesn't answer it,
call state_assumption and keep going.
When you stop before the work is finished, call hand_off.
After each finished change, call log_work in plain words.
The fuller version, with decisions and work in progress, is in the team agent instructions template. For Codex and Cursor in particular, see Codex, Cursor and other MCP clients.
What should you watch for?
- A personal token by mistake. If you weren't signed in to a team when you generated the token, your agent sees only what it creates. Pick the team in the header first, then generate a new token.
- A token in a commit. The repository set-up reads the token from the environment for exactly this reason. Never paste a token into a file you commit, and treat a team token like a password: anyone with it can read and edit the team's non-private pages as you.
- Claude Code not on the PATH. If the paste can't run
claude mcp add, it says so. The hooks still work; copy the MCP only line from the connect page to add the tools. - A settings file that isn't valid JSON. The paste leaves it alone and tells you. Fix the file and run the line again.
- Different folder names. If one person's clone sits in a folder with a different name, their sessions file their work under a different project. Keep the name the same.
- Old sessions. A session that was open before you connected won't have the hooks. Start a new one.
How do you know it's working?
- The connect page turns green. Below the command it shows "Waiting for your agent's first call" until something reaches Campfyrr with your token, then "Connected", with the tool that was called and when.
- The paste prints a brief. If it says the brief hook returned nothing, run it again or check the app is reachable.
- A new session opens with the team's state. Your own unfinished work and handoffs to you are at the top, under "Pick up where it was left".
- Opening a file brings an area brief. The first time a session touches, say,
lib/billing, it's told what's being changed there, the decisions to follow and what was tried. - Teammates can see your work. Your session's first request shows up as the goal of its work in progress on the project's Activity page.
- Codex answers from the record. Ask it to call
session_brief and it comes back with the team's state, not a blank.
What does Campfyrr automate once you're connected?
In Claude Code, with the one-paste or repository set-up, the hooks take care of the rest without anyone remembering:
- Session start: a brief of what changed since your last session, with your own unfinished work and handoffs to you at the top.
- First touch of an area: an area brief for that part of the code, named from its folder, such as "Billing · lib/billing": what's being changed right now and by whom, the decisions to follow, what changed this week, what was tried and dropped, and open assumptions.
- The first request: it becomes the one-line goal of the session's work in progress, visible to the team.
- Each turn: answers and corrections to the session's assumptions, and news of work it built on that has landed.
- Commit or pull request: the change's reasons, written for the reviewer, with a link for the description.
- Stop, compact or end: a handoff written from the session. The transcript itself isn't kept.
Codex, Cursor and other clients call the same tools by instruction. Campfyrr doesn't read or store your code, and secrets are stripped before any AI model sees text; the details are on the trust page.
Two minutes of set-up, once per person, and every session after that starts where the team is, not where it was.
The principle behind this: Keep moving: The work never waits, and neither do we.
Move as one at agent speed. Nobody waits. Add Campfyrr to Claude: it briefs every agent before it touches the code, keeps assumptions and handoffs where the team can see them, and never takes your code (what it keeps).