Collaboration & Messaging

#Collaboration & messaging

Group messaging, meetings, and video conferencing. Every connector follows the standard @toolset / @toolify shape.

Tip: read-only registration is usually enough for triage/notification agents. Use add_toolset(..., include_tags=["read"]) to drop write tools, or exclude_tags=["destructive"] to keep writes but drop the irreversible ones.

Agent-ready pattern. Across all four connectors, broad list / search / history tools return compact human-readable summaries by default with a response-local _ref field (channel_ref, message_ref, team_ref, space_ref, meeting_ref, etc.) and a small default page size (usually 10-25). Raw provider IDs are omitted unless you pass include_ids=True; pass include_metadata=False to get the raw provider response with every field. Write tools that target a single item (update_message, delete_message, send_chat_message, ...) accept either the raw ID string or the summary dict directly — the connector extracts the ID for you.

#SlackToolSet

SlackToolSet wraps the Slack Web API. It authenticates with a bot (xoxb-...) or user (xoxp-...) token and exposes identity, channel / IM lifecycle, messaging, threads, reactions, pins, files, reminders, and search.

Python

from maivn import Agent
from maivn_tools import SlackToolSet, register_connector

connector = SlackToolSet(token="xoxb-...")
agent = Agent(name="slack-agent", model="auto")
register_connector(agent, connector)

The connector forwards the token as a bearer credential. Scope enforcement happens on Slack's side; advertised scopes include channels:read, channels:history, chat:write, chat:write.public, reactions:read, reactions:write, pins:read, pins:write, files:read, files:write, search:read, users:read, users:read.email, users.profile:read, users.profile:write, reminders:read, reminders:write, emoji:read, team:read.

#Tools

Identity and presence. auth_test, user_lookup, list_users, get_user_profile, get_user_presence, set_user_status, team_info, emoji_list.

Channels and conversations. list_channels(types, exclude_archived, limit, cursor, include_metadata, include_ids), channel_info, channel_members, channel_history(channel, limit, cursor, oldest, latest, include_metadata, include_ids), thread_replies, create_channel, rename_channel, set_channel_topic, set_channel_purpose, join_channel, leave_channel, invite_to_channel, kick_from_channel, archive_channel, unarchive_channel, open_im.

Messages, pins, reactions. post_message, post_ephemeral, update_message, schedule_message, get_permalink, delete_message (destructive), add_reaction, remove_reaction, pin_message, unpin_message, list_pins.

Files. list_files, file_info, delete_file (destructive).

Search. search_messages(query, sort, sort_dir, count, page, include_metadata, include_ids), search_files.

Reminders. add_reminder, list_reminders, delete_reminder (destructive).

Every Slack response is checked for ok=false. The connector raises SlackApiError (a subclass of ProviderError) with the original Slack error string and response body in detail.

#Agent-ready behavior

list_channels, channel_history, and search_messages default to compact summaries with channel_ref / message_ref and the friendly fields an agent can quote (name, sender, text, ts, permalink). Raw Slack IDs (C0123, user IDs) are hidden — pass include_ids=True when a write tool needs them, or include_metadata=False to bypass summarization entirely. list_channels populates a name -> ID cache so write tools (post_message, update_message, delete_message, etc.) accept a friendly channel name, the raw ID, or the channel-summary dict from list_channels. update_message / delete_message / add_reaction similarly accept a raw ts string or a message-summary dict from channel_history.

Python

# Summaries — safe for final answers, ordinal refs can change between calls.
slack.search_messages("from:alice db migration")
# { "messages": [{"message_ref": "message_1", "username": "alice",
#                  "channel_name": "ops", "text": "...", "ts": "...",
#                  "permalink": "https://..."}], "total": 3, ... }

# Opt-in raw IDs (only when a follow-up tool needs them).
slack.search_messages("from:alice db migration", include_ids=True)

#Destructive tools

delete_message, delete_file, and delete_reminder are tagged destructive=True. They are not reversible — confirm with the user before invoking, or drop them at registration with exclude_tags=["destructive"].

#Pagination

Slack pagination uses opaque cursor values. The connector accepts a cursor argument on list_channels, list_users, channel_members, channel_history, and thread_replies, and returns the raw response_metadata.next_cursor in the response body. Use CursorPaginator from the runtime layer to walk multiple pages:

Python

from maivn_tools import CursorPaginator


def fetch(cursor):
    page = connector.list_channels(cursor=cursor)
    return page, page.get("response_metadata", {}).get("next_cursor") or None


paginator = CursorPaginator(fetch, items_key="channels")
all_channels = list(paginator.iter_items())

#MicrosoftTeamsToolSet

Microsoft Teams via Microsoft Graph v1.0.

Python

from maivn_tools import MicrosoftTeamsToolSet

connector = MicrosoftTeamsToolSet(token=token_cache)

Tools: list_joined_teams(include_metadata, include_ids), get_team, list_channels(team_id, include_metadata, include_ids), get_channel, create_channel, delete_channel (destructive), list_channel_messages(team_id, channel_id, top, include_metadata, include_ids), send_channel_message, list_channel_replies, reply_to_channel_message, list_members, add_member, remove_member (destructive), list_chats(top, include_metadata, include_ids), get_chat, list_chat_messages(chat_id, top, include_metadata, include_ids), send_chat_message.

#Agent-ready behavior

list_joined_teams, list_channels, list_chats, list_channel_messages, and list_chat_messages return compact summaries by default (team_ref, channel_ref, chat_ref, message_ref with friendly display names, content previews, sender, and created_at). Graph GUIDs are omitted unless include_ids=True; include_metadata=False returns the raw Graph response. Default page size is 25. list_joined_teams and list_channels populate name caches so get_team, get_channel, send_channel_message, and reply_to_channel_message accept a display name, a raw GUID, or the summary dict directly.

Python

teams.list_channel_messages(team_id="Engineering", channel_id="general")
# { "messages": [{"message_ref": "message_1", "sender": "Alice",
#                  "content_preview": "Build is green", ...}] }

#Destructive tools

delete_channel and remove_member are tagged destructive=True and remove user access / channel history. Filter them out with exclude_tags=["destructive"] for read-mostly agents.

#GoogleChatToolSet

Google Chat (Workspace) v1.

Python

from maivn_tools import GoogleChatToolSet

connector = GoogleChatToolSet(token=token_cache)

Tools: list_spaces(page_size, page_token, include_metadata, include_ids), get_space, create_space, delete_space (destructive), search_spaces(query, page_size, page_token, include_metadata, include_ids), list_messages(space, page_size, page_token, filter, include_metadata, include_ids), get_message, send_message, update_message, delete_message (destructive), list_members, add_member, remove_member (destructive), create_reaction, delete_reaction (destructive).

#Agent-ready behavior

list_spaces, search_spaces, and list_messages return compact summaries by default with space_ref / message_ref plus display name, space type, sender, and a 200-char text preview. Default page size is 25 (20 for messages). Provider resource names (spaces/AAA, spaces/AAA/messages/m1) are hidden unless include_ids=True; include_metadata=False returns the raw response. list_spaces and search_spaces populate a display-name cache so get_space, send_message, list_messages, etc. accept the friendly display name, the raw resource name, or the summary dict. update_message / delete_message / create_reaction accept the message dict from list_messages directly.

Python

google_chat.list_messages("Eng on-call")
# { "messages": [{"message_ref": "message_1", "sender": "Alice",
#                  "text_preview": "Pager fired again ...", ...}],
#   "next_page_token": null }

#Destructive tools

delete_space, delete_message, remove_member, and delete_reaction are tagged destructive=True. Confirm with the user before invoking — delete_space removes the entire message history.

#ZoomToolSet

Zoom Meetings + Webinars via the v2 REST API.

Python

from maivn_tools import ZoomToolSet

connector = ZoomToolSet(token=secrets["ZOOM_TOKEN"])

Tools: list_users, get_user, create_user, delete_user (destructive), list_meetings(user_id, type, page_size, next_page_token, include_metadata, include_ids), get_meeting, create_meeting, update_meeting, delete_meeting (destructive), list_meeting_participants, list_meeting_registrants, list_webinars, get_webinar, create_webinar, list_recordings, get_meeting_recordings, delete_meeting_recordings (destructive).

#Agent-ready behavior

list_meetings returns compact summaries by default with meeting_ref, topic, start_time, duration, and join_url and a default page size of 25. Raw Zoom meeting IDs are omitted unless include_ids=True; include_metadata=False returns the raw Zoom response. get_meeting, update_meeting, delete_meeting, list_meeting_participants, list_meeting_registrants, and get_meeting_recordings accept either a raw meeting ID or the meeting-summary dict from list_meetings(include_ids=True).

Python

zoom.list_meetings()
# { "meetings": [{"meeting_ref": "meeting_1", "topic": "Eng standup",
#                  "start_time": "2026-05-16T15:00:00Z",
#                  "duration": 30, "join_url": "https://..."}], ... }

#Destructive tools

delete_user, delete_meeting, and delete_meeting_recordings are tagged destructive=True. delete_meeting_recordings defaults to action="trash" (reversible until purged); action="delete" is irreversible. delete_user defaults to disassociate (keeps the user) rather than full delete. Confirm any of these with the user.