io.github.leshchenko1979/fast-mcp-telegram

Measured conclusion: Deep-tested. L2 outcome: tools_ok.

Multi-tenant Telegram gateway for AI agents — HTTP+stdio, 8 tools, MTProto User API

Add to your agent

Copy this into your MCP client config:

{
  "mcpServers": {
    "fast-mcp-telegram": {
      "url": "https://tg-mcp.l1979.ru/v1/mcp",
      "transport": "streamable-http"
    }
  }
}

Measured tools (8)

• search_messages_globally read-only

  Search all Telegram chats at once (not scoped to one chat). Comma-separated query terms; optional filters by date, chat kind, and public username. Results are ordered by ascending date-time (oldest first); limit selects the newest N. Success: message list and metadata dict. Global search ignores include_total_count. Full documentation: https://github.com/leshchenko1979/fast-mcp-telegram/blob/main/docs/Tools-Reference.md

  Parameters:

  • query (required) — string — Search terms, comma-separated for multiple terms (OR-style global search). Required.
  • limit — integer — Maximum messages to return (recommended 50 or less).
  • public — boolean — If true, prefer chats with a public username; if false, without. Does not apply to private DMs. Omit to skip this filter.
  • max_date — string — Inclusive maximum date filter (ISO 8601 date or datetime). Omit for no upper bound.
  • min_date — string — Inclusive minimum date filter (ISO 8601 date or datetime). Omit for no lower bound.
  • chat_type — string — Comma-separated chat kinds: private, bot, group, channel. Case-insensitive; extra spaces allowed.
  • auto_expand_batches — integer — Extra search batches to run when filters narrow results. Higher values may return more matches at the cost of latency.
  • include_total_count — boolean — If true, response may include total_count where supported (per-chat search; ignored for global search).

  Hints: read-only yes · destructive not declared · idempotent yes · open-world yes

• get_messages read-only

  Read or search messages in one chat: browse latest, search text, fetch by ids, or load replies to a message (comments, forum topics, threads). Use from_user to filter by sender (server-side, per-chat only). Use context to include neighboring messages and reply chains around each result. Use include_replies to fetch up to 5 direct replies per result. Do not combine message_ids with query or reply_to_id. Results are ordered by ascending date-time (oldest first) in every mode, including message_ids and nested context lists; limit selects the newest N. Success: messages, has_more, optional total_count and discussion fields. Full documentation: https://github.com/leshchenko1979/fast-mcp-telegram/blob/main/docs/Tools-Reference.md

  Parameters:

  • chat_id (required) — string — Target chat: numeric id (e.g. -100…), username without @, or 'me' for Saved Messages.
  • limit — integer — Maximum messages to return (recommended 50 or less).
  • query — string — Search within this chat only; comma-separated terms. Omit to browse latest or use message_ids / reply_to_id modes.
  • context — integer — Number of surrounding messages to include as context for each search result. 0 = disabled (default). 1-10 = include N messages before and N after each result. Also fetches the message being replied to and top replies (if include_replies=true). Requires chat_id. Disabled when result count exceeds cost-based caps.
  • max_date — string — Inclusive maximum date filter (ISO 8601 date or datetime). Omit for no upper bound.
  • min_date — string — Inclusive minimum date filter (ISO 8601 date or datetime). Omit for no lower bound.
  • from_user — string — Only return messages from this sender. Not a display-name or contact-name search — bare strings resolve like chat_id via get_entity (usernames are case-insensitive and may match an unrelated channel). Prefer @username, phone (+…), or numeric user id. Also accepts 'me', 'self', t.me URL, -100 prefixed id. Uses Telegram's native from_id server-side filter (per-chat search only).
  • message_ids — array — Exact message ids to fetch. Mutually exclusive with query and reply_to_id.
  • reply_to_id — integer — Anchor message id: channel post id, forum topic_id from get_chat_info, or a message id for direct replies. Use with thread_scope.
  • thread_scope — string — Only with reply_to_id. auto: full forum topic (topic_id) or channel comment thread via getReplies; else direct replies. full: nested branch under a message id (forum in-topic uses search window, not whole topic); supergroup threads use search top_msg_id. direct: immediate replies only.
  • include_replies — boolean — If true, fetch up to 5 direct replies per search result and attach as replies. Each result costs one API call (not batchable). Default: false.
  • auto_expand_batches — integer — Extra search batches to run when filters narrow results. Higher values may return more matches at the cost of latency.
  • include_total_count — boolean — If true, response may include total_count where supported (per-chat search; ignored for global search).

  Hints: read-only yes · destructive not declared · idempotent yes · open-world yes

• send_message destructive

  Send text and optional file attachments to a Telegram chat. Supports reply-to (including forum topics and channel discussion groups), parse_mode: classic markdown/html/auto (entities) or rich (Rich Message document; dialect auto-detected). parse_mode=rich cannot be combined with files. File attachments as http(s) URLs, local paths, or data: URIs. When files are provided, the message text becomes a caption. For channel posts with reply_to_id, automatically posts in the linked discussion group. schedule_date (ISO 8601, must be in the future) queues the message instead of delivering it now, so the result carries status='scheduled' plus scheduled and schedule_date. save_as_draft=true saves the text as the chat's draft instead of sending it (empty message clears the draft; user sessions only, bots cannot; text-only, so no files, no parse_mode='rich', no schedule_date). Success: dict with message_id, date, chat, text, status='sent', and sender info (rich messages also set rich=true and rich_format). Error: dict with ok=false and error string. Use send_message to create new messages; use edit_message to modify existing ones. Use send_message_to_phone when targeting a phone number instead of a chat_id. Full documentation: https://github.com/leshchenko1979/fast-mcp-telegram/blob/main/docs/Tools-Reference.md

  Parameters:

  • chat_id (required) — string — Target chat: numeric id (e.g. -100…), username without @, or 'me' for Saved Messages.
  • message (required) — string — Message text. When sending files, used as caption.
  • files — array — List of attachment URLs, local paths, or data URIs (one or more strings). data: URIs (data:<mime>;base64,<payload>) work in all server modes; local paths work in stdio mode only.
  • parse_mode — string — 'markdown'/'html'/'auto': classic entity formatting (auto detects). 'rich': Telegram Rich Message document; dialect auto-detected (known HTML tags outside code → rich HTML, else rich markdown). Default is 'auto'. parse_mode='rich' cannot be combined with files.
  • reply_to_id — integer — Telegram message id to reply to. For forums, topic root id; for channel posts, post id (may create a comment). Omit for a new top-level message.
  • save_as_draft — boolean — If true, save the text as the chat's draft instead of sending it (an empty text clears the existing draft). User sessions only — bot accounts cannot call messages.saveDraft. Cannot be combined with schedule_date, files, or parse_mode='rich'.
  • schedule_date — string — ISO 8601 instant at which Telegram should deliver the message, e.g. '2026-10-20T09:00:00Z' or '2026-10-20T12:00:00+03:00'. Must be in the future. On send: queues the message instead of delivering now. On edit: reschedules an already-queued message.

  Hints: read-only not declared · destructive yes · idempotent not declared · open-world yes

• edit_message destructive

  Replace the text of an existing message in a Telegram chat. Only works on messages sent by the authenticated account. Cannot edit media or other message attributes — text only. parse_mode: classic markdown/html/auto or rich (Rich Message; dialect auto-detected). schedule_date (ISO 8601) reschedules an already-queued message; it does not turn a delivered message into a scheduled one, and the result then carries status='scheduled'. Success: dict with message_id, date, chat, text, status='edited', and edit_date (rich messages also set rich=true and rich_format). Error: dict with ok=false and error string (e.g. message not found or not editable). Use edit_message to update a previously sent message; use send_message to create new ones. Full documentation: https://github.com/leshchenko1979/fast-mcp-telegram/blob/main/docs/Tools-Reference.md

  Parameters:

  • chat_id (required) — string — Target chat: numeric id (e.g. -100…), username without @, or 'me' for Saved Messages.
  • message (required) — string — Message text. When sending files, used as caption.
  • message_id (required) — integer — Message id in this chat to edit (from get_messages or Telegram).
  • parse_mode — string — 'markdown'/'html'/'auto': classic entity formatting (auto detects). 'rich': Telegram Rich Message document; dialect auto-detected (known HTML tags outside code → rich HTML, else rich markdown). Default is 'auto'. parse_mode='rich' cannot be combined with files.
  • schedule_date — string — ISO 8601 instant at which Telegram should deliver the message, e.g. '2026-10-20T09:00:00Z' or '2026-10-20T12:00:00+03:00'. Must be in the future. On send: queues the message instead of delivering now. On edit: reschedules an already-queued message.

  Hints: read-only not declared · destructive yes · idempotent yes · open-world yes

• find_chats read-only

  Find users/groups/channels by name, username, or phone. Comma-separated usernames are searched in parallel and results are merged round-robin. Global search (query required) searches all Telegram; with min_date, max_date, or filter, search uses dialog list or a named filter; include_peers filters use last-activity from GetPeerDialogs; flag-based filters use dialog list dates. Success: dict with key chats (list of chat objects). Full documentation: https://github.com/leshchenko1979/fast-mcp-telegram/blob/main/docs/Tools-Reference.md

  Parameters:

  • limit — integer — Maximum chats to return (recommended 50 or less).
  • query — string — Name, username (no @), phone (+country…), or comma-separated usernames for batch lookup. Example: 'alice,bob,charlie'. Required for global search unless you use min_date/max_date or folder alone.
  • folder — string — Telegram folder name (case-insensitive exact match after normalization). In Telegram's UI these are called folders; internally they are "dialog filters" — saved filter presets that group chats by custom criteria (pinned, unread, business, etc.). See Filters-vs-Folders.md for the technical distinction.
  • public — boolean — If true, prefer chats with a public username; if false, without. Does not apply to private DMs. Omit to skip this filter.
  • max_date — string — Inclusive maximum date filter (ISO 8601 date or datetime). Omit for no upper bound.
  • min_date — string — Inclusive minimum date filter (ISO 8601 date or datetime). Omit for no lower bound.
  • chat_type — string — Comma-separated chat kinds: private, bot, group, channel. Case-insensitive; extra spaces allowed.

  Hints: read-only yes · destructive not declared · idempotent yes · open-world yes

• get_chat_info read-only

  Load profile and metadata for one user, bot, group, or channel. Success: info dict; forum chats may include topics up to topics_limit; user targets may include common_chats up to common_chats_limit. Full documentation: https://github.com/leshchenko1979/fast-mcp-telegram/blob/main/docs/Tools-Reference.md

  Parameters:

  • chat_id (required) — string — Target chat: numeric id (e.g. -100…), username without @, or 'me' for Saved Messages.
  • topics_limit — integer — Max forum topics to list when the chat is a forum.
  • common_chats_limit — integer — Max common groups to list for user targets.

  Hints: read-only yes · destructive not declared · idempotent yes · open-world yes

• send_message_to_phone destructive

  Send to a phone number: may create a temporary contact, then send text or files. Supports parse_mode: classic markdown/html/auto or rich (Rich Message; dialect auto-detected). parse_mode=rich cannot be combined with files. schedule_date (ISO 8601, future) queues the message instead of delivering now; save_as_draft=true saves the text as the draft in that chat instead of sending. Success: send result plus contact_was_new / contact_removed when applicable. Full documentation: https://github.com/leshchenko1979/fast-mcp-telegram/blob/main/docs/Tools-Reference.md

  Parameters:

  • phone_number (required) — string — E.164 phone number with country code, e.g. +1234567890 (must be on Telegram).
  • message (required) — string — Message text. When sending files, used as caption.
  • files — array — List of attachment URLs, local paths, or data URIs (one or more strings). data: URIs (data:<mime>;base64,<payload>) work in all server modes; local paths work in stdio mode only.
  • last_name — string — Last name when creating a temporary contact.
  • first_name — string — First name when creating a temporary contact.
  • parse_mode — string — 'markdown'/'html'/'auto': classic entity formatting (auto detects). 'rich': Telegram Rich Message document; dialect auto-detected (known HTML tags outside code → rich HTML, else rich markdown). Default is 'auto'. parse_mode='rich' cannot be combined with files.
  • remove_if_new — boolean — If true, delete the contact after send when it was created only for this send.
  • save_as_draft — boolean — If true, save the text as the chat's draft instead of sending it (an empty text clears the existing draft). User sessions only — bot accounts cannot call messages.saveDraft. Cannot be combined with schedule_date, files, or parse_mode='rich'.
  • schedule_date — string — ISO 8601 instant at which Telegram should deliver the message, e.g. '2026-10-20T09:00:00Z' or '2026-10-20T12:00:00+03:00'. Must be in the future. On send: queues the message instead of delivering now. On edit: reschedules an already-queued message.
  • reply_to_msg_id — integer — Reply to this message id in the target chat after resolve.

  Hints: read-only not declared · destructive yes · idempotent not declared · open-world yes

• invoke_mtproto destructive

  Low-level Telegram API (MTProto) invoke for methods not wrapped by other tools. Dangerous methods require allow_dangerous=true. Success: API result dict or normalized error. PII and credential-shaped fields (phone, access_hash) are dropped from a successful result by default; pass include_sensitive=true for the raw payload. A bare message id needs a chat binding: requests with no peer field (messages.GetMessages, messages.DeleteMessages) are refused, because a bare id resolves against an arbitrary dialog. Use channels.GetMessages or messages.GetHistory, which carry the binding. messages.GetHistory cannot address a forum topic (no thread_id/top_msg_id in the schema, and channels.GetHistory does not exist) -- use messages.Search with top_msg_id, or the high-level get_messages with reply_to_id. Full documentation: https://github.com/leshchenko1979/fast-mcp-telegram/blob/main/docs/Tools-Reference.md

  Parameters:

  • method_full_name (required) — string — Telegram API method, e.g. "messages.GetHistory" or "users.GetFullUser" (normalization applied).
  • params_json (required) — string — JSON object string of TL parameters as in Telegram API docs; nested TL uses "_": "typeName" discriminator.
  • resolve — boolean — If true, resolve string/int peer-like fields to TL Input* entities before invoke.
  • allow_dangerous — boolean — If false, destructive methods (e.g. deletes) are blocked. Set true only when intended.
  • include_sensitive — boolean — If true, return the raw result including PII and credential-shaped fields (phone, access_hash). Default false drops them.

  Hints: read-only not declared · destructive yes · idempotent not declared · open-world yes

Publisher website Repository

Measured by MCPYP probing. Untested fields are omitted, never estimated. Listing is always free; MCPYP never proxies calls or stores user tokens.