On a team, every developer's agent ends up knowing something different. One person's Claude Code was told about the new billing rules last week. Another's Cursor never was. A third pasted an old version of the release checklist into Codex. Ask all three the same question and you get three answers, and no way to tell which one is current.
Shared context means every agent on the team reads from the same place: the same knowledge, the same written procedures, kept current in one spot.
Why each agent drifts
Agents get context from what is near them: the repository, a rules file, and what the person in front of them pastes. All three are per person or per repository.
- Rules files are per repository and per tool. Claude Code reads
CLAUDE.md, Codex readsAGENTS.md, Cursor reads its project rules. Company knowledge copied into each one drifts. - Pasted context is per person. It lives in one conversation and is gone after it.
- Nobody owns the copies. When a decision changes, the doc gets updated. The copies in rules files and old chats don't.
What shared context looks like
One source the whole team's agents read, with three properties:
- One copy. Knowledge lives once, kept current from where your team writes it.
- Every agent reads it the same way. Claude Code, Cursor, Codex and the rest, over MCP.
- Each person signs in as themselves. Agents act as the person using them, so you can see who wrote what.
This is what Noetix is built to be. The rest of this guide sets it up.
Set it up for a team
Build one brain
One person sets up the workspace: start from your company's website, then connect Google Drive, Notion or Slack. A workspace has exactly one brain, so there is no question of which one to connect to.
Invite the team
An admin copies the invite link from Settings › Workspace. Everyone who joins asks and adds to the same brain.
Connect each agent
Every person adds Noetix to the agent they use, with the same address. Pick yours:
Claude Code
Anthropic's terminal coding agent.
claude mcp add --transport http noetix https://mcp.noetix.dev/mcpCodex CLI
OpenAI's coding agent.
[mcp_servers.noetix]
url = "https://mcp.noetix.dev/mcp"
auth = "oauth"Then sign in once:
codex mcp login noetixCursor
The editor. Global config, or per-project if you'd rather scope it.
{
"mcpServers": {
"noetix": {
"type": "http",
"url": "https://mcp.noetix.dev/mcp"
}
}
}Windsurf
Cascade reads its own config file, with its own key name.
{
"mcpServers": {
"noetix": {
"serverUrl": "https://mcp.noetix.dev/mcp"
}
}
}Claude Desktop
The desktop app. No file to edit, it has a UI for this.
- Open Settings → Connectors
- Choose Add custom connector
- Paste this as the URL
https://mcp.noetix.dev/mcpOpenClaw
Self-hosted assistant gateway. Slack, Discord, WhatsApp, Telegram.
openclaw mcp add noetix \
--url https://mcp.noetix.dev/mcp \
--transport streamable-http \
--auth oauthThen sign in once:
openclaw mcp login noetixHermes Agent
Nous Research's headless runtime, usually on a server you own.
hermes mcp add noetix --url https://mcp.noetix.dev/mcp --auth oauthThen sign in once:
hermes mcp login noetixAny other MCP client
Cline, Zed, VS Code or your own. If it speaks MCP, it connects.
{
"mcpServers": {
"noetix": {
"type": "http",
"url": "https://mcp.noetix.dev/mcp"
}
}
}Sign in once
Each person approves their agent in the browser the first time it connects. No shared key, nothing to rotate when someone leaves.
Write procedures down once
Shared knowledge answers "what". Teams also need shared "how": how to ship, how to review a pull request, how to record a decision. Write each one as a skill. Every connected agent finds the skill that matches its task and follows it, so a release done from Cursor and one done from Claude Code follow the same steps.
Members can draft a skill; an admin publishes it before any agent sees it. See Skills.
Keep it current
Shared context is only useful while it is current.
- Sources update on their own. Google Drive and Notion files you imported are checked for changes and replaced. Your website is checked every week.
- Agents write back. When an agent learns something worth keeping, it can edit a page or add a line to the Notes page. Every version is kept, so nothing an edit does is lost.
- Decisions get their own pages. When the team changes its mind, the old decision is marked replaced, not deleted, so agents see both and know which is current.
- Gaps are listed. When an agent's question finds nothing, it lands on Knowledge gaps, so the team can see what to write next.
Who sees what
Everything in the brain is visible to everyone in the workspace, and to their agents. There are no per-source permissions yet, so connect only what your whole team may read.
Members can ask, add and edit pages. Admins also connect sources, publish skills and approve removals. See Workspace and roles.
One line in each repository
So every agent reaches for the brain first, add the same short section to each
repository's rules file (CLAUDE.md for Claude Code, AGENTS.md for Codex):
## Company knowledge
Our company's knowledge is in Noetix, connected as the `noetix` MCP server.
- Before guessing why something is the way it is, `recall` it.
- For a task with a house procedure, use `search_skills`, then `use_skill`.
- If the brain has nothing on a question, `flag_gap` once with the question.