How agents read company Telegram accounts
Each person connects their own agent over MCP, and it reads only their granted accounts.
From a grant to a read
The agent works inside the access its person already has.
- 01
Connect your own agent
Sign in from any MCP client with OAuth, or use a token that expires.
Why?
With OAuth there is no token to copy. A token is the fallback for a client without it, and it shows its expiry date.
- 02
It sees your grant, nothing more
The connection is yours and capped by your grant. Owners and admins can revoke it.
Why?
Revoke the grant, or remove the person from the team, and the connection ends with it. An owner or admin can connect across the organization; a member connects only the accounts granted to them.
- 03
It reads, and it is recorded
It lists accounts, checks status and reads chats. Each read goes into your audit log.
Why?
The audit entry names the person whose agent made the call. An agent can also ask for a summary of a period, or for the chats with open questions.
What an agent cannot do
- It cannot send, edit or delete a message. Sending stays a person's act.
- It never sees a Telegram login code: codes are masked before any agent reads.
- It takes no orders from a chat: message text reaches it marked as untrusted.
- It cannot call without limit: each connection has caps per minute and per day.
Questions
Which AI agents can connect?
Any MCP client that signs in with OAuth, or your own agent with a token.
Do we have to host a Telegram MCP server ourselves?
No. Nicegram Business hosts it, and the Telegram sessions stay on our backend.
Why?
A self-hosted Telegram MCP server runs on a session you keep on your own machine. Here no session is handed to a device, so the company can end access without asking for one back.
What happens when someone leaves?
Revoke their grant or remove them, and their agent's connection ends too.
Why?
A connection belongs to one person and one granted account, so nobody inherits it. The account and its archive stay with the company.
Can we see what an agent read?
Yes. Each read is in your audit log, naming the person behind it.
Why?
Owners and admins read the log. Answers from the built-in assistant are recorded the same way.
Which model does the agent use?
Your own. It runs on your model and uses none of the included AI.
See it with your own agent
Thirty minutes with a person. A reply within one working day.