What MCP access is
MCP (Model Context Protocol) is the standard AI agents use to connect to external services. Metriox supports it, so an agent can reach your project's analytics on its own: look through events, compute a metric, find a particular user.
Instead of "export a report and paste it into the chat", you just ask the agent, and it fetches the data itself.
Questions an agent can answer on its own
- "How many unique users messaged the bot in the last week?"
- "Find payment error events from yesterday and show which step they happen on"
- "Which bots are connected to the project, and how many events has each sent?"
- "Are we running past the plan's event limit?"
What an agent can and cannot do
This is the most important section on the page. Read it before you hand out a token.
A token is limited to one project
A token is minted for a specific project and works only with that one. If you have several projects, each needs its own token. An agent physically cannot reach another project's data, even if the person who minted the token has access to it.
A token has exactly the rights of whoever minted it
A token acts with the role its owner holds in the project right now. A token minted by a Writer can do what that Writer can; a token minted by an admin can do what an admin can, including starting a broadcast.
There used to be a hard Writer ceiling here, and no token could start a broadcast at all. It was removed deliberately, and the price is worth stating plainly: a leaked admin token can do anything an admin can, and an owner's token reaches billing and project deletion. So mint a token from the role the job needs, not from the highest role available to you.
What no token can do at any role: mint or revoke other MCP tokens, or exceed the plan's limits.
What an agent can change
Thirty-one tools, listed in the reference. Twenty only read. Eleven change data: dashboards (create, rename, replace cards, delete) and broadcasts (create a draft, edit it, prepare it, send a test message, start it, recall it).
metriox_test_send_campaign delivers a message to the chat you name, and metriox_start_campaign sends
it to every recipient. Starting requires admin rights; a test send does not.
There is an undo, but a partial one: metriox_cancel_campaign (admin rights too) stops the run and
recalls messages already delivered — but only those still inside Telegram's deletion window.
Anything past it stays with the recipient for good.
If an agent should only prepare broadcasts and leave sending to a person, mint the token from a Writer: it can build the draft and cannot start it.
Rights are checked on every request
A token acts on behalf of the person who minted it, and that person's role is resolved on every request. So if they are removed from the project or demoted, the token loses access immediately, with no revocation needed.
Minting a token
- Open the Metriox app
- Go to Project settings and find MCP tokens
- Click Mint a token and give it a name you will recognise (for example,
claude-code-laptop) - Copy the token: it is shown once
- Pick your program in the dialog to get a ready-to-use connection command
Next: setting up your client.
We store only an HMAC fingerprint of the token, so showing it again is impossible. If you lose it, revoke that token and mint a new one.
It grants access to the project with the rights of whoever minted it. Do not commit it to a repository or paste it into files that end up in git. Where the client supports it, use an environment variable or a prompt at startup: the setup page shows the safer option for each program.
Lifetime and revocation
A token is minted for 90 days. An expired token stops working, so mint a new one.
You can revoke a token at any time in the same MCP tokens section; revocation takes effect on the agent's next request. The list shows when each token was last used, which is the quickest way to tell which one nobody needs any more.