关于 AwrAIter — Telegram channels
AwrAIter 是面向 Telegram 频道管理者的远程 MCP 服务,通过 AwrAIter 账户操作频道。它结合频道分析、基于来源的内容草拟、内容规划和定时发布,不使用 Telegram userbot。
主要功能
- 01查看频道统计数据与表现洞察
- 02搜索频道帖子和已连接来源
- 03依据频道规范草拟并检查帖子
- 04创建内容计划并查找空闲时段
- 05安排发布、发布或取消帖子
- 06管理品牌简介与频道经验记录
适用场景
- 01查看近期帖子的触达率和互动表现
- 02根据已抓取的频道来源草拟帖子
- 03围绕可用时段规划后续帖子
- 04通过 MCP 客户端安排频道内容发布
MCP 相关
通过远程 Streamable HTTP MCP 服务,为账户授权的 Telegram 频道提供 29 个工具。连接使用 OAuth 2.1 或个人访问令牌,令牌可设为只读或完整权限。
配置
远程服务无需本地安装。选择你使用的客户端,复制配置或使用一键安装。
https://awraiter.ai/mcpclaude mcp add --transport http ai-awraiter-telegram https://awraiter.ai/mcp{"mcpServers":{"ai-awraiter-telegram":{"type":"http","url":"https://awraiter.ai/mcp"}}}在 Cursor 中打开{"servers":{"ai-awraiter-telegram":{"type":"http","url":"https://awraiter.ai/mcp"}}}在 VS Code 中打开{"mcpServers":{"ai-awraiter-telegram":{"type":"http","url":"https://awraiter.ai/mcp"}}}{"mcpServers":{"ai-awraiter-telegram":{"serverUrl":"https://awraiter.ai/mcp"}}}命令在终端中运行;JSON 粘贴到对应客户端的 MCP 配置后重启。需要授权的服务会在首次连接时提示登录。
工具
README 原文English
来自项目 READMEREADME 暂无中文译文,以下为英文原文;中文介绍见上方。
AwrAIter MCP Server
A remote MCP server for the Telegram channels you administer: read analytics, draft posts from your own scraped sources, build content plans, schedule and publish — from inside ChatGPT, Claude, or any MCP-capable client.
Not a userbot. No phone number, no login code, no .session file. The server acts through your AwrAIter account and reaches only the channels of the team you authorize.
Русская версия · Connection guide · Tool reference · Team and invites · awraiter.ai
Server
| URL | https://awraiter.ai/mcp |
| Transport | Streamable HTTP |
| Authentication | OAuth 2.1 (PKCE, dynamic client registration) or a personal access token |
| Registry name | ai.awraiter/telegram |
| Availability | Every plan, including the free tier |
Quick start
You need an AwrAIter account (sign in with Telegram) and at least one connected channel.
Claude (web / desktop)
Settings → Connectors → Add custom connector, paste the URL, press Connect, and confirm access for the team you want.
https://awraiter.ai/mcp
Claude Code
claude mcp add --transport http awraiter https://awraiter.ai/mcp
Agent skill (Claude Code, Codex and other agents)
The AwrAiter_skills repository ships a skill with the full tool contract, authentication, the draft → check → preview → schedule workflow and the guardrails an agent should follow:
# Claude Code — as a plugin
claude plugin marketplace add MediaZione/AwrAiter_skills
claude plugin install awraiter@awraiter
# Codex and any other agent with skills
npx skills add MediaZione/AwrAiter_skills --skill awraiter
ChatGPT
AwrAiter is not in the ChatGPT plugin catalog yet — add it as your own MCP server: Settings → Security and login → Developer mode, then chatgpt.com/plugins → +, paste the URL, keep OAuth and press Connect. In a chat, pick it from the tools menu. The ChatGPT/Codex desktop form has no OAuth option; connect it with a token instead (see below).
Clients that take a JSON config
{
"mcpServers": {
"awraiter": {
"type": "http",
"url": "https://awraiter.ai/mcp"
}
}
}
Clients without OAuth (token auth)
Create a token in the panel under MCP connection — it is shown once, together with ready-to-paste snippets (Claude Code command, JSON config, cURL, Node.js) that already contain it — and send it as a header:
Authorization: Bearer YOUR_TOKEN
Step-by-step instructions for Claude, Claude Code, ChatGPT, Codex, Gemini CLI and Qwen Code are in docs/connect.md; ready-to-paste config files are in examples/.
Tools
29 tools. The ones that change something are marked ✍️ — everything else is read-only.
| Tool | What it does |
|---|---|
list_teams | Teams you belong to; marks the active one |
switch_team ✍️ | Switch the active team (membership is verified server-side) |
list_channels | Channels owned by the active team |
get_channel | One channel's profile: name, chat id, language, description |
get_channel_stats | Subscribers, reach, ER and dynamics over 1–90 days |
get_channel_insights | What works for this audience, scored against the channel's own 30-day median |
review_posts | How recent posts did against the channel's own posts at the same age: above, on par or below |
get_recent_posts | Most recent published posts, newest first |
search_channel_content | Semantic search over the channel's own posts and source material |
search_sources | Search the channel's bound sources |
draft_from_sources | Freshest scraped sources not yet used in a post |
deep_analyze ✍️ | Deep AI analysis of channel performance — the one tool that spends the team's AI credits |
check_uniqueness | Is this draft a near-duplicate of existing posts or a copy of a source? |
score_post | Data-grounded quality signal for a draft, specific to this channel |
preview_post | Exact message that would be sent — title, body, media, formatting |
create_post_from_text ✍️ | Save a ready-to-publish draft from text the assistant wrote, with photos, videos or music tracks by link |
schedule_post ✍️ | Publish a draft now, or schedule it for a given time |
cancel_scheduled_post ✍️ | Cancel a scheduled post by its plan id |
get_content_plan | The channel's plan configuration and its slots |
create_content_plan ✍️ | Create a content plan for a channel |
find_free_slot | Earliest slot in the plan with no post yet |
get_channel_memory | The channel's brand brief, banlist and learnings journal |
set_brand_brief ✍️ | Replace the channel's brand/tone guidance |
add_banlist_entry ✍️ | Add a forbidden word to the channel's banlist |
remove_banlist_entry ✍️ | Remove a word from the banlist |
add_channel_learning ✍️ | Record a hypothesis or a proven learning about what works in the channel |
remove_channel_learning ✍️ | Remove a journal entry |
set_assist_mode ✍️ | How much the assistant coaches you: quiet, normal or coach |
get_playbook | Step-by-step operator playbook for a workflow |
What each tool returns, and the exact scope it needs, is in docs/tools.md.
Authentication
OAuth 2.1 with PKCE and dynamic client registration; redirect URIs are restricted to HTTPS or HTTP loopback, authorization codes are single-use, and refresh tokens rotate. A token is bound to one user, one team and one scope.
Personal access tokens come in two scopes:
| Scope | Allows |
|---|---|
read | Read-only tools; every ✍️ tool is refused server-side |
full | Everything the account can do in the panel |
Tokens are listed, and can be revoked, in the panel under MCP connection.
Security model
| Userbot-style Telegram MCP servers | AwrAIter | |
|---|---|---|
| What you hand over | Phone number, login code, session file | An OAuth grant to your own account |
| What it can reach | Everything your account can — DMs, groups, private material | Only the channels of the team you authorized |
| How it publishes | As you | Through the panel's own service admin, with the same permission checks as the UI |
| If the token leaks | Your whole Telegram account | Revoke the token in the panel; nothing else is exposed |
Every write goes through the same channel-level access control as the panel, and rate limits apply per token. The full model, and how to report a vulnerability, are in SECURITY.md.
About AwrAIter
AwrAIter is a panel for people who run Telegram channels: weekly content plans, AI drafts in your channel's tone, scheduled publishing, and analytics for everything published — growth, reach, ER/ERR, per-post performance, channel comparison, export to XLSX or Google Sheets. One channel is free forever; paid plans start at €9/month. See awraiter.ai.
Support and legal
- Questions and issues: awraiter.ai/support
- Terms of Service · Privacy Policy
- Operated by Mediazione Italia S.r.l. Payments are handled by Stripe; we never see card numbers.
- Your channel content stays yours — we do not train models on it unless you explicitly opt in (off by default).
License
Documentation and examples in this repository are released under the MIT License. The AwrAIter service itself is proprietary.
评论
一个开放协议,规定了 AI 应用如何以统一方式连接外部工具与数据源。客户端负责与模型对话,MCP 服务负责提供工具、资源和提示模板,两者通过标准消息通信,同一个服务可以被不同客户端复用。