tg-mcp
Read-only MCP server for Telegram chats and channels that provides digest summaries, message search, and action items.
README
tg-mcp
Read-only Telegram digest MCP server for selected Telegram chats and channels.
Current shape
- Node.js 22 ESM service.
- Express + MCP Streamable HTTP endpoint at
/mcp. - REST/OpenAPI fallback under
/tg-mcp/apiand/tg-mcp/openapi.json. - MongoDB storage.
- GramJS-based Telegram sync CLI.
- Optional read-only Telegram slash bot.
- Read-only MCP tools:
list_sourcesget_sync_statusget_daily_digestget_period_summarysearch_telegram_messagesget_message_contextget_action_itemsget_source_summary
Local development
npm ci
npm run cli -- setup-env
npm test
npm start
Health:
curl http://127.0.0.1:3010/health
REST/OpenAPI fallback:
curl http://127.0.0.1:3010/tg-mcp/openapi.json
curl "http://127.0.0.1:3010/tg-mcp/api/sync/status"
curl "http://127.0.0.1:3010/tg-mcp/api/digest/daily?timelineLimit=80"
curl "http://127.0.0.1:3010/tg-mcp/api/sources/<sourceId>/summary?date=2026-07-09"
curl "http://127.0.0.1:3010/tg-mcp/api/digest/daily?sourceQuery=project"
curl "http://127.0.0.1:3010/tg-mcp/api/digest/daily?refresh=true"
curl "http://127.0.0.1:3010/tg-mcp/api/search?query=release"
If APP_AUTH_TOKEN is set, pass Authorization: Bearer <token> for REST and MCP calls.
Telegram setup
Fill these values in .env or /srv/tg-mcp/shared/.env:
TELEGRAM_API_ID=
TELEGRAM_API_HASH=
TELEGRAM_SESSION_FILE=/srv/tg-mcp/shared/sessions/telegram.session
ALLOWED_SOURCE_IDS=
Create or update the env file:
npm run cli -- setup-env --set TELEGRAM_API_ID=<api_id> --set TELEGRAM_API_HASH=<api_hash>
For the VPS layout:
export TELEGRAM_API_ID=<api_id>
export TELEGRAM_API_HASH=<api_hash>
npm run cli -- setup-env --production --env-path /srv/tg-mcp/shared/.env --from-env TELEGRAM_API_ID --from-env TELEGRAM_API_HASH
Existing secrets are preserved unless you pass a new value with --set or --from-env. The command writes mode 0600, creates a backup before overwriting an existing file, and generates APP_AUTH_TOKEN when it is missing.
List available sources:
npm run cli -- login
npm run cli -- doctor --telegram
npm run cli -- list-sources
Save the source list into MongoDB, then enable selected chats/channels:
npm run cli -- refresh-sources
npm run cli -- find-sources project
npm run cli -- select-source "Project Alpha" --tag work
npm run cli -- sync
Useful variants:
npm run cli -- disable-source <id>
npm run cli -- set-source-tags <id> --tag work --tag project-x
npm run cli -- enable-source <id> --tag work
npm run cli -- db-sources --include-disabled
npm run cli -- sync --source-id <id> --limit 100
npm run cli -- backfill --days 7 --limit 1000
Alternatively, put selected source ids into ALLOWED_SOURCE_IDS; env selection overrides DB-enabled sources during sync.
login is interactive and writes the Telegram session file. Later commands reuse that session and should not prompt unless Telegram requires reauthorization.
Check readiness:
npm run cli -- doctor
npm run cli -- doctor --telegram
doctor returns machine-readable checks plus nextSteps with the next safe commands for the current setup state. doctor --telegram also performs a non-interactive authorization check with the existing session file.
Admin operations
Authenticated admin endpoints are available for operations that should not be exposed as MCP tools:
curl -X POST http://127.0.0.1:3010/admin/sync \
-H "Authorization: Bearer <APP_AUTH_TOKEN>" \
-H "Content-Type: application/json" \
-d '{"sourceIds":["<sourceId>"],"limit":100}'
Use {"backfillDays":7} to bypass the incremental cursor for a historical import. The endpoint uses the existing Telegram session file and never prompts.
Background sync
The HTTP service can run a safe background sync loop after the Telegram session file exists:
TELEGRAM_SYNC_ENABLED=true
TELEGRAM_SYNC_INTERVAL_SECONDS=300
TELEGRAM_SYNC_ON_START=true
If credentials, session, or selected sources are missing, the worker logs a warning and waits for the next interval. It never prompts from the systemd service.
Normal sync is incremental: each source tracks lastSyncedMessageId and later runs request only newer Telegram messages. backfill --days N intentionally bypasses that cursor for historical imports.
Check data freshness:
curl "http://127.0.0.1:3010/tg-mcp/api/sync/status?staleAfterHours=24"
The MCP tool get_sync_status exposes the same source freshness state to ChatGPT so it can say when data is missing or stale before summarizing.
Digest cache
Daily, period, and source summaries are cached in tg_digests. The cache key includes the period, timezone, source filters, timeline options, and selected source sync state, so a later Telegram sync naturally invalidates stale summaries.
Use refresh=true in REST calls, or refresh: true in MCP tool arguments, to force recomputation from stored messages.
Optional Telegram slash bot
The HTTP service can also run a small read-only Telegram bot for quick checks against the same selected data:
TELEGRAM_BOT_ENABLED=true
TELEGRAM_BOT_TOKEN=<bot token from BotFather>
TELEGRAM_BOT_ALLOWED_CHAT_IDS=<your chat id or comma-separated ids>
TELEGRAM_BOT_TIMEZONE=Europe/Chisinau
Supported commands:
/digest_today [source]
/digest_week [source]
/search <query>
/actions [source]
/sources [query]
TELEGRAM_BOT_ALLOWED_CHAT_IDS is optional, but recommended. The bot never sends Telegram messages on behalf of the synced user account; it only replies with digests/search results from MongoDB.
VPS quick deploy
The target server already has Apache, MongoDB, and bundled Node.js.
Expected layout:
/srv/tg-mcp/
repo.git/
releases/
shared/
.env
logs/
sessions/
node/
current -> releases/<release>
Deploy code:
ops/deploy.sh
Install systemd service:
sudo ops/install-systemd-service.sh
Install Apache /mcp proxy:
sudo ops/install-apache-proxy.sh
Install log rotation for /srv/tg-mcp/shared/logs/*.log:
sudo ops/install-logrotate.sh
That installer exposes:
https://celticspear.com/mcp
https://celticspear.com/tg-mcp/openapi.json
https://celticspear.com/tg-mcp/api/...
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。