MCP vs CLI: two ways to plug your AI into Atlas, and how to set up each
A big part of a coach's week goes to looking things up.
Who missed sessions. What an athlete logged on Tuesday. Who runs out of program next week. Moving Thursday's squats to Friday because someone has a work trip. Every one of those is a few clicks, and a few clicks times a whole roster is your Sunday night.
If you already pay for an AI assistant, it should be doing that part. The problem is your assistant can't see your coaching platform. So you copy and paste, and at that point it's faster to do it yourself.
Atlas has no built-in AI. Nothing gets pushed on you. If you want AI, you bring the one you already use, and it does only what you allow.
I keep an eye on the other coaching platforms. As of November 14, 2026, Atlas is the only coaching platform with an MCP connector. CoachRx doesn't have one [12]. CoachLogik and CoachRx build their own AI in, with Copilot and RxBot [13]. TrainHeroic says its outside connections are closed [14]. TeamBuildr says it doesn't offer a public way for outside tools to connect [16]. TrueCoach gives you a Zapier key on Pro and up [15], and Everfit has a way for outside tools to connect for its Enterprise customers [17]. Everyone's going a different direction, and that's fine. Mine is letting you plug in whatever assistant you already like.
Our MCP connector just went live, so there are now two ways to hook your assistant up: the MCP and the CLI. Here's what each one is, who should pick which, and how to set up both.
What each one is, in plain words
The MCP is a connector you add to your AI app. You paste one Atlas address into Claude, ChatGPT, Grok or Gemini, sign in to Atlas, approve it, and done. Nothing to install. After that you just ask in plain language in the same chat you already use, on the web, and in some apps on your phone too.
The CLI is a small tool your AI runs on a computer. It's for AI agents that can do work on a computer, like Claude Code, Codex inside ChatGPT, or Cursor. The agent installs it, you approve the sign-in, and from then on the agent can run Atlas for you from that computer.
Both reach the same Atlas. Same athletes, same programs, same permissions. Anything the docs list your assistant doing through one, it can reach through the other. The difference is where you're working when you ask.
Which one should you pick
| MCP connector | CLI | |
|---|---|---|
| Where you use it | Chat apps: Claude, ChatGPT, Grok, Gemini | Agents that work on a computer: Claude Code, Codex, Cursor, Gemini CLI or Antigravity, OpenCode, Muse, Grok Bot |
| Install | None. Paste one address | The agent downloads it for you |
| Phone | Yes in Claude and Gemini, once you add it on the web | Only through an agent that runs on its own computer |
| Sign-in | Approve it in the Atlas console | Approve a short code in the Atlas console |
| Changes | The chat app usually asks you before a change | The Agent Skill tells the agent to show you changes first |
| Best for | Quick questions and everyday edits | Big jobs, files on your computer, building your own programming setup |
Pick the MCP if
- You live in a chat app and don't want to install anything.
- You want to ask from your phone between sessions. In Claude and Gemini you add the connector on the web once and it shows up in the phone app.
- Most of what you need is quick: who checked in sore, what someone hit on their snatch, move a day, add a note to the week.
Pick the CLI if
- You already use an agent like Claude Code, Codex or Cursor.
- The job is big. Turning a full spreadsheet program into an Atlas block, setting up a new athlete from start to finish, or reviewing the whole roster at the end of a block. The CLI runs on the computer where your files already are, so the agent can open your spreadsheet right where it sits and work through it step by step.
- You want to build your own programming setup. The Agent Skill (more on that below) can capture how you program and turn it into a programming bot that writes in your style.
You don't have to choose forever. You can run both. Each connection shows up separately in Atlas, so you can see and cancel each one on its own.
How to set up the MCP connector
1. Copy the Atlas connector address.
https://atlas.meetcal.app/mcp
It's also in the Atlas console under Settings, then Integrations.
2. Add it to your AI app as a custom connector.
As of November 2026, here's what each app's own help page says. These menus move around, so if a step looks different, check the linked page.
- Claude: on claude.ai or the Claude desktop app, open Customize, then Connectors, choose Add, then Add custom connector. Give it a name, paste the address, and follow the sign-in. Connectors you add there also work in the Claude phone apps. On Claude's Free plan you get one custom connector. If you use Claude Code signed in with your Claude account, connectors you add on claude.ai show up there too.
- ChatGPT: on the web, open Plugins, choose the plus button, then Add custom MCP server. Give it a name, paste the address, and follow the sign-in. OpenAI says custom servers are web only, not in the phone app. It also says full support, including making changes, is for ChatGPT Business and Enterprise/Edu workspaces, and Pro accounts can only connect to read and look things up. Check your plan before you set this up.
- Grok: go to grok.com/connectors, choose New Connector, then Custom, paste the address, and follow the sign-in.
- Gemini: on your computer, go to gemini.google.com, open Settings, then Connected Apps, then Add a custom app. Paste the address, choose Next, and follow the sign-in. Google says this needs a personal Google account, in the US, for someone 18 or over, with Keep Activity on, and is only available in English. You add it on the web. Once it's added, it works in the Gemini phone app too. If you don't see Connected Apps under Settings, Google says to open Personal Intelligence, then Connected Apps. Atlas doesn't list Gemini as a known app yet, so read the next step carefully.
3. Approve it in Atlas.
Your AI app sends you to an Atlas page. Check the name of the app asking and the site it returns to. Atlas marks Claude, ChatGPT and Grok as known apps. Approve any other app only if you just started connecting it yourself.
Then pick three things:
- Which club. Each connection is for one club.
- What it can do. More on this below.
- How long it lasts. When it expires, your app asks you to approve it again.
Approve, and you're connected. There's no password or key to copy anywhere. You sign in and approve it, that's it.
4. Start with a question that only reads.
"Who missed sessions this week?" is a good first one. If the answer matches what you see in Atlas, you're set.
How to set up the CLI
1. Let your agent install it.
This is the easy way. Open Claude Code, Codex, Cursor or whichever agent you use, and tell it:
"Install the Atlas CLI from https://atlas.meetcal.app/docs/cli and sign me in with read and write."
It downloads the right version for that computer (Mac, Windows or Linux) and hands you a link and a short code.
2. Approve the code in Atlas.
Open the link. If you're not signed in to the Atlas console, it asks you to sign in first. The page is titled Authorize Atlas CLI and shows the code. Make sure it matches what your agent showed you. Then pick the club, what it can do, and how long it lasts: 30, 90 or 180 days, or no expiry. For an AI agent, 30 days is a good pick.
The access gets saved in your computer's own password storage. Your agent uses it to run Atlas, and it never sees your Atlas password.
3. Give your agent the Atlas Agent Skill.
The skill is a set of instructions that teaches your agent how Atlas works and how to work with you. Copy it from the Agent Skill card in Settings, then Integrations, or from atlas.meetcal.app/docs/agent-skill, and save it where your agent keeps instructions:
- Claude Code: a SKILL.md file in a skills folder.
- Codex: AGENTS.md in your working folder.
- Cursor: a project rule or a skill.
Or paste it in and say "save this as a skill for working with Atlas."
The skill tells the agent to start with what you want, and to show you changes before saving them.
4. Doing it by hand instead.
If you'd rather install it yourself, the CLI page in the docs has every download plus a section for people who have never opened a terminal. One heads up for Windows: that build isn't signed yet, so Windows may warn you before it runs. The docs show how to check the file first.
Which agents can run the CLI
Atlas's docs say the agent has to be able to run commands on a computer: Claude Code, Codex, Cursor, or a hosted agent with its own computer. From the list I recommend to coaches, that's:
- Claude: Claude Code.
- ChatGPT: Codex, inside ChatGPT.
- Gemini: Gemini CLI or Antigravity.
- Cursor.
- OpenCode.
- Muse and Grok Bot: these run on their own computer in the cloud, so your laptop doesn't have to stay open.
A regular chat window can't run the CLI. That's what the MCP connector is for.
Staying in control
Handing an AI your athletes' data should make you careful. It's built so you decide what it touches.
- It only gets what you pick. Reading your roster and programs, making edits, health data, messages, and deleting things are all separate permissions. Atlas never pre-selects deleting, messaging or health data. You turn those on yourself if you want them.
- One club at a time. A connection only works in the club you picked when you approved it.
- It asks before changing things. With the MCP, the chat app usually asks before each change. Gemini's help says it requires you to confirm any change [11]. ChatGPT's help says it may ask, depending on the app's permissions and the action [9]. One exception: Anthropic says that when you run Research in Claude, it can use connector tools on its own without asking again, so turn off Atlas's write tools before you run Research [6]. With the CLI, the Agent Skill tells the agent to show you changes for review before saving.
- Athletes see what changes. Program edits and messages show up for your athletes, so read what your assistant proposes before you say yes.
- Oura and FatSecret data only come through if you give it health data access. Atlas reads those live and doesn't store them.
- You can cut it off anytime. Every connection shows up in Settings, then Integrations, named for the app or agent. Cancel it there and it stops working. If a coach leaves your club, their connections stop working too.
Here's a simple way to think about what to allow:
- Reviewing your roster and analytics: read only.
- Editing programs, notes and templates: add create and edit.
- Oura and FatSecret: add health data.
- Reading messages: add read messages. Sending messages or replying to comments: add send messages.
Start small. You can always approve a new connection with more later.
What to actually ask for
Once it's connected, talk to it like you'd talk to an assistant coach.
Everyday stuff
- "Who missed sessions this week?"
- "Who checked in sore this morning, and who skipped their pre-lift questionnaire?"
- "Look at my roster and tell me who runs out of program this week, and who has an empty week coming."
- "Move Thursday's squats to Friday for this athlete and add a note to the week."
- "What's still open? Unanswered comments, things I said I'd do, pain or schedule notes."
- "Export this block as a PDF so I can send it."
Reviewing a block
- "Did the last block move the snatch? Which lift stalled? Did readiness track the misses?"
- "Compare this block to the last one for this athlete."
- "What time of day does this athlete lift best?"
Video reviews
- "Read my last video review and show me the cues." Approve them, and it can add them to that lift's notes for the weeks ahead.
New athletes
- "Set up this new athlete: create the profile, load their PRs, work out training weights, assign my template, and set up a meet plan for their next meet."
Programming, if you want it
This part is optional. If you want help writing programs, the Agent Skill starts by learning how you coach. You can talk it through, or give it programs you've already written in Google Sheets, PDFs or other sites, and it saves your methodology so it writes in your style instead of a generic program. Then give it a spreadsheet and it checks one block and shows it to you before it creates anything in Atlas. Take that far enough and you've built your own programming bot.
That's where the CLI shines, since the agent works with your files right on your computer.
If you've never used an AI assistant, you don't need any of this. Everything above can be done by hand in Atlas.
Try it
Start Free, add the connector to the AI app you already use, and ask it who missed sessions this week. If setup trips you up, email me at maddisen@meetcal.app.
Sources
All facts checked November 14, 2026.
- Atlas Docs, Connectors: https://atlas.meetcal.app/docs/connectors
- Atlas Docs, Atlas CLI: https://atlas.meetcal.app/docs/cli
- Atlas Docs, Agent Skill: https://atlas.meetcal.app/docs/agent-skill
- Atlas Docs, Getting Started (permissions and access): https://atlas.meetcal.app/docs
- Atlas AI Assistants: https://atlas.meetcal.app/features/ai-assistants
- Claude Help Center, custom connectors using remote MCP: https://support.claude.com/en/articles/11175166-getting-started-with-custom-connectors-using-remote-mcp
- Claude Help Center, connectors on web, desktop and mobile: https://support.claude.com/en/articles/11176164-use-connectors-to-extend-claude-s-capabilities
- Claude Code Docs, MCP: https://code.claude.com/docs/en/mcp
- OpenAI Help Center, connecting custom MCP servers in ChatGPT: https://help.openai.com/en/articles/12584461-developer-mode-and-full-mcp-connectors-in-chatgpt-beta
- xAI Docs, Grok connectors: https://docs.x.ai/grok/connectors
- Gemini Apps Help, custom connected apps: https://support.google.com/gemini/answer/17209137
Competitor facts checked November 14, 2026.
- CoachRx Roadmap: https://www.coachrx.app/roadmap
- CoachLogik, Why we built CoachLogik's Copilot the way we did (Aug 28, 2026): https://coachlogik.com/blog/why-we-built-coachlogik-s-copilot-the-way-we-did
- TrainHeroic Support, Can I customize the TrainHeroic mobile app or access an API?: https://support.trainheroic.com/hc/en-us/articles/18156691093773-Can-I-customize-the-TrainHeroic-mobile-app-or-access-an-API
- TrueCoach Help, Zapier integration: https://help.truecoach.co/en/articles/8688374-zapier-integration
- TeamBuildr Pricing (FAQ): https://www.teambuildr.com/pricing
- Everfit Help, Everfit API integration guide: https://help.everfit.io/en/articles/15176278-everfit-api-integration-guide