AwrAiter

Model Context Protocol

Telegram MCP server for ChatGPT and Claude

Plan, publish and measure the channels you own — from inside your assistant's chat window. One hosted connector, OAuth or a token, 25 tools, works on the free plan.

Connector URL — the same for every client

https://awraiter.ai/mcp

Streamable HTTP. OAuth 2.0 with a consent screen, or a personal access token you mint yourself in the panel. 120 requests per 60 seconds per token.

01 — Connect

Three steps, once

  1. 01

    Add the connector

    In your client, add a custom MCP server and paste https://awraiter.ai/mcp. Transport is Streamable HTTP.

  2. 02

    Sign in with Telegram

    OAuth opens an AwrAiter page: sign in and confirm which team the assistant may act for. Clients without OAuth take a token from awraiter.ai/mcp-connect as Authorization: Bearer ….

  3. 03

    Ask it something

    The tools appear in the client's tool list. “Show last week's statistics for my channel.”

02 — Clients

Per client

Claude — web & desktop

Settings → Connectors → Add custom connector → paste the URL → Connect. Sign in with Telegram and confirm the team.

Claude Code

claude mcp add --transport http awraiter https://awraiter.ai/mcp — it walks you through OAuth on first use.

ChatGPT & Codex

In the browser, Settings → Connectors offers OAuth. The desktop/Codex form has no OAuth: set type Streamable HTTP and add the header Authorization: Bearer pat_…. Leave the “token environment variable” fields empty — a token typed there leaves the connector with no tools.

Cursor, Gemini CLI, Qwen Code, anything else

Any MCP-capable client works against the same URL with a Bearer token in the headers.

03 — Tools

What the assistant can do — all 25 tools

Read tools work with any scope. ✍️ marks tools that change something: they need OAuth or a full-scope token. Only deep_analyze spends AI credits — everything else is free on every plan.

Teams & channels

list_teamsYour teams, with the active one marked.
switch_teamSwitch the active team.
list_channelsChannels of the active team.
get_channelOne channel's profile — name, language, description.

Analytics & insights

get_channel_statsThe dashboard: subscribers, reach, ER, dynamics over 1–90 days.
get_channel_insightsWhat works for this audience: top and flop posts scored against the 30-day median, plus statistically-backed patterns — best day, posting window, length, question at the end — once the channel has 20+ scored posts.
deep_analyzeDeep AI analysis of one channel, or a comparison of several. Runs on our servers and spends AI credits.
get_recent_postsThe channel's latest published posts.
search_channel_contentSemantic search across the channel's own past posts and sources.

Writing & publishing

draft_from_sourcesThe freshest scraped sources not yet used in a post — raw material to draft from.
search_sourcesSearch the sources bound to the channel.
create_post_from_textSave a finished post — your assistant's text, plus images and video by URL — as a draft. ✍️
preview_postExactly what will be sent: title, body, media.
score_postScore a draft against the channel's own history before it goes out.
check_uniquenessNear-duplicate check against the team's posts and sources.
schedule_postPublish a draft now or at a chosen time. ✍️
cancel_scheduled_postRoll a scheduled post back. ✍️

Content plans

get_content_planThe channel's plan and its slots.
find_free_slotThe earliest free slot in the plan.
create_content_planCreate a plan for a channel. ✍️
get_playbookA step-by-step operator playbook: analyze, create_plan, fill, improve, rewrite.

Channel memory

get_channel_memoryThe channel's brand brief and banlist.
set_brand_briefSet free-form brand and tone guidance the AI follows. ✍️
add_banlist_entryAdd a forbidden word to the channel's list. ✍️
remove_banlist_entryRemove a word from the channel's list. ✍️

04 — Example prompts

Example prompts

Paste one into your assistant once the connector is on.

  • Compare my three channels over the last 30 days and tell me which content works best where.
  • Take the freshest unused source for @mychannel, write a post in the channel's tone, score it, and if the score is good — schedule it for the next free slot.
  • What are the statistically best day and time to post on my channel?
  • Build a content plan for next week: 5 posts, alternating news and evergreen topics.
  • Check this draft against my last 100 posts for near-duplicates, then preview exactly what would be sent.

05 — How it differs

Not the same thing as a Telegram userbot server

Most Telegram MCP servers hand the assistant your personal account over MTProto so it can read your chats. This one never touches your account or your correspondence. It works on the channels where you are an administrator, and on what happens to them: the plan, the publishing, the numbers.

What differsUserbot MCP serversAwrAiter
What it connectsYour personal Telegram accountThe channels you administer
What it seesYour chats and correspondenceYour channels' posts, plans and statistics
CredentialsYour phone number and a session fileOAuth sign-in or a revocable token
PublishingMessages from you, as youScheduled posts into the channel, with preview and scoring
AnalyticsNoneGrowth, ER, ERR, per-post stats, channel comparison
HostingUsually self-hostedHosted — nothing to run

06 — FAQ

Questions people actually ask

Do I have to give it my Telegram password or a session file?

No. You sign in with Telegram on our page, or paste a token you minted in the panel. There is no phone number, no code, no session file — and nothing that would let an assistant read your personal chats.

Can the assistant publish without asking me?

Only if you let it. A read-only token can read analytics and content and cannot publish, edit or cancel anything; publishing needs OAuth or a full-scope token. Every write goes through the same permission checks as the panel, including per-channel access for team members.

Does it work on the free plan?

Yes. The connector is included on every plan, including the free one — one channel, free forever. Only deep_analyze spends AI credits.

How is this different from telegram-mcp servers that read my chats?

Those give the assistant your personal account over MTProto. AwrAiter gives it the channels you administer: content plans, scheduled publishing, ER/ERR analytics and post-level insights. Different job, different data, different risk.

Which clients are supported?

Anything that speaks MCP over Streamable HTTP: Claude web and desktop, Claude Code, ChatGPT and Codex, Cursor, Gemini CLI, Qwen Code and others. OAuth where the client offers it, a Bearer token where it does not.

Are there rate limits?

120 requests per 60 seconds per token. Agents running in a tight loop should back off.

Connect a channel and ask your assistant about it

One channel is free forever, no card. The connector is included.