Atarim MCP: Connecting Claude, Codex, and Other Coding Agents
Connect Claude, Codex, or any MCP-compatible assistant to your Atarim workspace in three steps: copy the Server URL, add it to your assistant, and authorize the connection.
If your developers work inside an AI coding assistant — Claude, Codex, or something similar — Atarim MCP connects that assistant directly to your workspace. Once connected, it can read tasks, look up sites, and even trigger approved actions on its own, so a client request captured in Atarim can turn into working code without anyone manually copying details across. Setup is short: copy your Server URL from Atarim, add it to your assistant, and approve a one-time browser prompt — the sections below walk through each part, including the exact setup for Claude, Codex, and other MCP-compatible tools.
Step-by-Step Guide
What This Actually Does
Think of it as a direct line between Atarim and your developer’s AI assistant. Normally, if a client leaves a request in Atarim, someone has to read it, understand it, and go explain it to whatever tool they’re coding in. With Atarim MCP connected, the assistant can just look the information up itself — what site it’s about, what the task says, what’s already been done. “MCP” stands for Model Context Protocol, an open standard that a growing number of AI tools support for exactly this kind of connection — Atarim is simply one of the things they can connect to.
Concretely, once connected, your assistant can:
- Read and manage tasks — pull up a task’s full content, change its status or priority, assign it, tag it, or break it into subtasks, all without anyone switching over to Atarim to do it manually.
- See every site and client in the workspace — list projects, look up a specific site’s details, screenshots, or pages, and check who a client is before starting work.
- Work with your automations — list existing workflows, check past runs, and even trigger one on demand — useful for a developer who wants to kick off a scheduled workflow early rather than waiting.
- Trigger an AI design review on a specific page or graphic directly from the conversation.
- Make real changes on a connected WordPress site through the same bridge the rest of Atarim uses — not just read information, but actually execute WordPress actions your assistant has access to.
That last point is the one worth being precise about: connecting your assistant this way doesn’t loosen any of Atarim’s existing safeguards. If something your assistant tries to do would normally need your sign-off in the Approval Queue, it still does — the tool that runs gated actions is explicitly built to “never decide on its own” and only proceed once you’ve clearly said yes or no in the conversation. Connecting MCP changes how the request reaches you, not whether it still needs your approval.
| Without MCP | With MCP connected |
|---|---|
| Someone reads the client’s task, then explains it to the developer | The assistant reads the task directly, in the conversation |
| Developer switches to Atarim to check status, tags, or history | The assistant looks it up without switching tools |
| Approval Queue still applies to anything gated | Approval Queue still applies — MCP changes the request path, not the safeguard |
Where to Turn It On
- In Atarim, from the main dashboard, select Settings.
- Under Connected Apps, select Atarim MCP.
- Under Server URL, select the copy icon at the right of the field. A message confirms Server URL copied to clipboard. The address is
https://api.atarim.io/v1/mcp. - Switch to Claude and open Settings, then Connectors.
- Select Add at the top right, then Add custom connector.
- In the first field, enter a name:
Atarim. - In the second field, paste the Server URL you copied.
- Leave OAuth Client ID and OAuth Client Secret empty — Atarim registers the client automatically.
- Select Add.
- A browser page opens headed Connect Claude to Atarim, naming the workspace. If you aren’t signed in to Atarim, you’ll sign in first and return here.
- Select Connect. The button shows Connecting…, then the page confirms You’re connected.
- You’re returned to Claude automatically. If not, select Close this tab.
- Back in Claude, confirm Atarim now appears in your connector list.
- Test it — ask Claude which Atarim workspaces you have access to. You should see your own workspaces named back, not a generic answer.










What It Can and Can’t See
Nothing extra. Whatever your Atarim account already has access to is exactly what a connected assistant can see and act on — nothing more. There’s no separate permissions screen to configure here; it simply follows the same access rules your account already has. If you want to limit what an assistant can touch, the way to do that is the same as anywhere else in Atarim — adjust what your own account has access to.
And on top of that access scoping, anything that would normally be gated behind your approval — like an install on a client’s site — stays gated. Your assistant can surface that a request needs a decision, but it can’t make that decision for you.
The Browser Prompt You’ll See
After you add the Server URL to your assistant, expect one extra thing to happen the first time:
- Your assistant opens a browser tab pointing at Atarim.
- You’ll see a summary of what the connection grants access to.
- Select Connect to authorize it — this is a standard, secure login step (OAuth), not an error.
- Close the tab. Your assistant is now connected.
This happens automatically as part of adding the URL in the steps below — you don’t need to go looking for it, just don’t be surprised when a browser window pops up mid-setup.
Choosing Your Setup Path
All three paths use the same Server URL — the difference is only in how you hand it to your assistant:
| Assistant | Method | Where to look |
|---|---|---|
| Claude Code (CLI) | One-line terminal command | Terminal |
| Claude Desktop | Paste URL into Settings, or edit a config file | Settings → Connectors / MCP servers |
| Codex | Terminal command, or edit the config file directly | Terminal, or Codex’s settings file |
| Other MCP-compatible tools | Paste URL into that tool’s own connector settings | Look for “Connectors,” “Integrations,” or “MCP servers” |
You only need to set this up in whichever assistant a given developer actually uses day to day — there’s no need to configure every tool.
Connecting Claude
Claude Code (command-line):
Add Atarim:
claude mcp add --scope user --transport http atarim https://api.atarim.io/v1/mcp
Start Claude Code:
claude
Manage servers (typed inside Claude Code, not the terminal):
/mcp
List servers:
claude mcp list
Remove Atarim:
claude mcp remove atarim --scope user
Check it’s installed:
claude --version
Claude Code
- Open a terminal.
- Run the following, using the Server URL copied from Atarim:
claude mcp add --transport http atarim https://api.atarim.io/v1/mcp - To make Atarim available in every project rather than only the current one, include
--scope user:claude mcp add --scope user --transport http atarim https://api.atarim.io/v1/mcp - Start a new Claude Code session.
- Enter
/mcpto open the list of MCP servers. - Select atarim, then choose Authenticate.
- Complete the approval step in the browser tab that opens.
- Return to the terminal. Atarim now shows as connected.





















Claude Desktop:
Atarim isn’t listed in Claude’s connector directory, so it’s added as a custom connector.
- In Atarim, copy the Server URL from Settings → Atarim MCP.
- In Claude, open Settings, then Connectors.
- Select Add, then Add custom connector.
- In the first field, enter a name:
Atarim. - In the second field, paste the Server URL:
https://api.atarim.io/v1/mcp - Leave OAuth Client ID and OAuth Client Secret empty. Atarim handles this automatically.
- Select Add.
- Complete the approval step described above.
- Atarim now appears in your connector list, and its tools are available in conversations.
Connecting Codex
- Open a terminal.
- Run the command below, using your actual Server URL from the Atarim MCP settings page.
codex mcp add atarim --url https://api.atarim.io/v1/mcp
If you prefer editing the settings file directly instead of using that command, add these lines to it:
[mcp_servers.atarim]
url = "https://api.atarim.io/v1/mcp"
Other MCP-compatible tools
- Open that assistant’s own settings.
- Look for a Connectors, Integrations or MCP servers section. MCP is an open standard, so most AI coding assistants support this even without Atarim-specific instructions.
- Paste in the Server URL from Atarim.
- If asked for a connection type, choose remote or HTTP rather than a locally-run option — Atarim’s server is on the web, not on your computer.
- Complete the approval step.











Once connected, the server identifies itself as Atarim, so that’s the name to look for in your assistant’s server list.onOther MCP-compatible tools
Confirming it worked
- Check that your assistant lists Atarim as connected. In Claude Code, enter
/mcp; in Codex, runcodex mcp list. - Ask a question that confirms both the connection and what it can reach:Which Atarim workspaces do I have access to?You should see the workspaces your Atarim account already has — and nothing else.
- Then try a real lookup:List the open tasks on [site name] in Atarim.

If the assistant answers generally without naming real workspaces or tasks, it isn’t reaching Atarim. Check that it lists Atarim as connected, and that you completed the approval step rather than closing the browser tab early.
What your assistant can access
Nothing extra. A connected assistant can see and act on exactly what your own Atarim account can — no more. There’s no separate permissions screen to configure; the connection follows the access rules your account already has.
Each person connects with their own Atarim account, so everyone’s assistant is limited to their own access. If you want to limit what an assistant can reach, adjust what that person’s Atarim account can reach.
Anything that would normally need your approval stays gated. Your assistant can tell you a request needs a decision, but it can’t make that decision for you.
Working with WordPress sites
Reading and managing Atarim data works for any site in your account. Acting on a website itself — editing content, changing settings — requires the Atarim plugin installed and connected on that site.
Your assistant identifies a site by its Atarim record rather than its web address, so it will usually list your sites first to find the right one.
Installing or activating a plugin always needs your approval. When your assistant tries one of these actions, it pauses and asks you to approve or decline before anything happens on the site. It won’t decide on your behalf
Disconnecting
Disconnect from within your assistant:
- Claude: Settings → Connectors, then remove the Atarim connector.
- Claude Code:
claude mcp remove atarim - Codex: remove the Atarim entry from your MCP server list or configuration file.
Removing the connector there ends that assistant’s access to Atarim.
FAQs
Can more than one person on my team connect their own assistant?
Yes. The Server URL is the same for everyone — each person adds it to their own assistant and signs in with their own Atarim account. Every connection is limited to that person’s own access.
Does this replace the WordPress plugin or the Chrome extension?
No. The plugin is what makes website actions possible — your assistant reaches a site through the Atarim plugin installed on it. Without the plugin, the connection still works for Atarim data such as tasks, comments and sites, but can’t act on the website itself. The Chrome extension is unrelated and unaffected.
What can my assistant actually see?
Only what you can. The connection acts as you, using your existing Atarim permissions. Connecting doesn’t grant new access to you or anyone else.
Can my assistant change things, or only read them?
It can make changes — creating and updating tasks, adding comments, and on plugin-connected sites, editing the site itself. Actions that install or activate plugins always pause for your approval first.
Conclusion
Setup takes a few minutes per developer: copy the Server URL, add it to their assistant, approve the one-time browser prompt, then confirm it actually works with a real question. Once that's done, it fades into the background — your developers just ask their assistant about a site or task, and it already knows, with every gated action still routing through your Approval Queue exactly as before.
Tips & best practices
- Set this up individually for each developer's own assistant rather than trying to share one setup across the team.
- Always do the final "ask it something real" check — don't assume it's working just because it shows as connected.