Launch campaigns
from your own AI agent

Connect the platform to your client with a single setup — and set tasks in plain words. The agent will pick sources, collect the audience, write the dialogue flow and check the campaign before launch.

One command in the terminal Where to get a key →

$ claude mcp add --transport http intelgram https://api.intelgram.ai/mcp --header "Authorization: Bearer igrm_ВАШ_КЛЮЧ"

{
  "mcpServers": {
    "intelgram": {
      "url": "https://api.intelgram.ai/mcp",
      "headers": { "Authorization": "Bearer igrm_ВАШ_КЛЮЧ" }
    }
  }
}

{
  "servers": {
    "intelgram": {
      "type": "http",
      "url": "https://api.intelgram.ai/mcp",
      "headers": { "Authorization": "Bearer igrm_ВАШ_КЛЮЧ" }
    }
  }
}

[mcp_servers.intelgram]
url = "https://api.intelgram.ai/mcp"
http_headers = { Authorization = "Bearer igrm_ВАШ_КЛЮЧ" }

{
  "mcpServers": {
    "intelgram": {
      "serverUrl": "https://api.intelgram.ai/mcp",
      "headers": { "Authorization": "Bearer igrm_ВАШ_КЛЮЧ" }
    }
  }
}

{
  "mcpServers": {
    "intelgram": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://api.intelgram.ai/mcp",
               "--header", "Authorization:${IGRM_KEY}"],
      "env": { "IGRM_KEY": "Bearer igrm_ВАШ_КЛЮЧ" }
    }
  }
}

The key is issued in your dashboard, in the API section. You tick the rights yourself — the agent cannot do more than you allowed.

agent + intelgram

Getting started

Three steps to your first task

There is nothing to install on your side: the server runs with us, the client reaches it over the network.

Issue a key

In your dashboard, in the API section. Tick the sections and rights the agent needs — audience, flows, campaigns.

Add the server

Pick your client on the first screen and copy the setup. The platform's tools show up right after you connect.

Say who you want to reach

In plain words. The agent will ask for what is missing with ready answer options and get to work.

Your first task

Say it like this

Ready-made wordings for the whole path — from finding people to an assembled campaign. Click any of them to copy and send it to the agent as is.

You do not have to be precise: whatever the agent is missing it will ask about itself, one question at a time and with answer options.

What it can do

What the agent does on its own

Twenty-two tools — the whole path from the first question to an assembled campaign.

Finds sources

Builds key phrases from your description and uses them to search Telegram for chats and channels with comments.

find_sources · evaluate_sources

Picks by the numbers

Looks at members, activity, whether a channel has a chat, and the labels — and explains its choice with figures, not with a chat's name.

evaluate_sources

Collects the audience

Queues parsing jobs for the chosen sources and watches the progress while people fill the list.

collect_audience · get_audience_task_progress

Labels and filters people

Creates variables, fills them with AI from your question about a person, and marks the ones worth writing to.

create_audience_variable · fill_audience_variable · mark_peers

Writes the dialogue flow

First it shows the plan in words — stages, the target action, answers to objections — and builds the graph only after you agree.

build_flow

Tests the flow on itself

Runs conversations several turns deep — interest, an objection, a refusal, an empty reply — and fixes what it finds before showing it to you.

test_flow

Guards the wording

Checks messages for wordings that get accounts reported, and warns you before sending rather than after a ban.

review_messages

Assembles the campaign

Brings the audience, the flow and the accounts together and says what is missing for the start. You confirm the launch.

assemble_campaign · list_accounts

Knows how it should be done

Its own knowledge base: why accounts burn, what the first message should look like, how to handle objections, when to close a dialogue.

list_knowledge · read_knowledge

Before you start

What you need to have

A key with rights

Rights are granted per section and separately for reading and writing. The key can be revoked in the dashboard at any moment — the agent's access disappears at once.

Tokens on the balance

You pay for the platform's work: finding sources, parsing chats, AI calls. The prices are the same as in the dashboard — the agent shows an estimate before starting.

An account package

Audience jobs and the outreach itself run from rented accounts. Without an active package the agent will say so plainly.

Questions

The short answers

An open protocol that gives an AI agent access to external tools. We host the server — you only add it to your client and set tasks in words: the agent calls the platform's methods itself and works out their order.

With any that speaks MCP over HTTP. The first screen holds ready setups for Claude Code, Cursor, VS Code, Codex CLI, Windsurf and Claude Desktop — pick yours and copy it. The rest need the same two things: the address https://api.intelgram.ai/mcp and an Authorization header with your key.

Only what the key's rights allow. Give read access and it can look but not change. The campaign launch stays with you either way: the agent brings it to an assembled state and shows what is blocking the start.

The server itself is free and comes with access to the platform. You pay for the work — finding sources, parsing chats, AI calls — with tokens from your balance at dashboard prices. Before expensive operations the agent shows an estimate and asks.

No. The server runs on our side, the client reaches it over the network. Updates land for everyone at once — there is nothing to reinstall.

Yes, the dashboard is not going anywhere: the same audiences, flows and campaigns. The agent is a way to walk the path faster and not skip steps, not a separate version of the platform. Everything it creates shows up in the dashboard with links.

Launch campaignsfrom your own AI agent