slack

name	slack
description	Use when searching or interacting with Slack messages, channels, or users. Also covers drafting messages in Andrey's voice.

Slack

Safety Rules

When using the Slack MCP server, follow these rules to avoid triggering Slack's anti-scraping detection (which revokes tokens and force-signs out):

• NEVER call channels_list or any bulk channel/user enumeration tool.
• Use #channel-name to reference channels — the server resolves them lazily via single API calls.
• Explore and cross-reference Slack data freely, but do it through targeted queries (search, history, replies), not bulk listings.
• Space out Slack API calls when making several in a row.

Search

• from:me goes in search_query, NOT in filter_users_from
• filter_users_from expects a username like @john or user ID, not from:me
• Date filters use YYYY-MM-DD format, not natural language like "this week"
• Use filter_in_channel with #channel-name or channel ID

Limitations

• users_search and conversations_unreads are not available (browser token)
• Never bulk query the Slack API (e.g. listing all users or all channels) — this will flag and revoke the user token. Paginating is fine but be mindful of volume and add sleeps between requests.
• The channel parameter in all Slack tools is called channel_id, not channel. Using channel will error.

Andrey's Slack Voice

When drafting Slack messages on Andrey's behalf, match this voice. GitHub issue comments stay detailed/formal — these notes only apply to Slack.

General style

• Slack collapses API-posted paragraphs over ~338 characters with "Show more" (native app doesn't do this). Since we post via the API, keep each paragraph under ~300 chars. Use \n\n to break longer messages into multiple paragraphs, and keep each one under the limit. This is a dead giveaway the message wasn't typed by hand.
• Casual and conversational, like texting a coworker
• Normal apostrophes in contractions: don't, can't, I'm, I'll, it's, haven't, doesn't, won't, didn't
• Usually capitalizes "I" but not always
• No markdown formatting (no backticks, bold, headers)
• Never use em dashes (—) — they're a dead giveaway for AI text. Use commas, periods, or just rewrite the sentence instead.
• Uses "y'all", "gonna", "lemme know", "atm"

Tone by context

• DMs/team chat: Very casual, stream-of-consciousness, sometimes witty/self-deprecating humor
• Cross-team requests: Friendly, provides context/links first, then the ask. Not overly formal but polite.
• Quick replies: "Thanks", "Thank you", "Sure thanks for the heads up", "interesting"
• Updates/standups: Doesn't use emoji bullet points, typically brief

Humor & personality

• Self-deprecating, witty asides
• Uses "haha", "hahaha" (not "lol")
• Emojis: :sweat_smile:, :grimacing:, :shrug:, :thinking_face:, :grin:, :hugs:

Technical messages

• Leads with context/investigation before the ask
• "I think maybe what happened was..."
• "seems like", "looks like", "I'm honestly not too sure why"
• Inline links, not formatted lists
• Provides evidence (curl output, jq commands, etc.)

When asking for help from other teams

• Explains the situation naturally, not bullet-pointed
• References PRs/issues inline
• Friendly but direct, doesn't over-apologize