MCP App Store
by KeenEthics
Data & Analytics
Vibewatch icon

Vibewatch

by BRANDON JAMES MARSHALL

Overview

Vibewatch is community intelligence for web3 teams. It reads your community across Discord, Telegram, X, Bluesky, and more, scores the vibe, writes daily highlights and lowlights, and sends a weekly report. Connect once and Claude can read your Vibewatch organization's data. Every tool is read-only. Ask Claude to: • Tell you the vibe in your community this week, with the breakdown by platform • Explain a sentiment dip using the daily trend and the messages behind it • Search messages by platform, date, sentiment score, author, or keyword • Summarize your latest weekly report, including what's worth addressing • Pull recent daily highlights and lowlights, with links to the source messages • Compare your community's vibe with the market backdrop: the Fear & Greed Index and the assets you track • Check the public Stacks ecosystem index across opted-in projects Sign in with your Vibewatch account and pick the organization Claude can read. An organization owner or admin turns on MCP access in Vibewatch settings. It's available on every plan.

Tools

get_daily_insights

ChatGPT
Get daily rolling insights (highlights and lowlights). Args: days: Number of days to retrieve (1-7, default: 3) Returns daily highlights and lowlights with source citations, composite sentiment score, and message counts. An empty list may mean data collection is paused for this organization (trial or subscription ended) and older data was already removed.

get_market_context

ChatGPT
Get market sentiment context (Fear & Greed Index, tracked assets). Args: days: History depth in days (1-90, default: 30) Returns current and historical Fear & Greed values, tracked token and stock info, the org benchmark index, and a sentiment-vs-market comparison time series.

get_organization

ChatGPT
Get info about the organization this API key is scoped to. Returns slug, name, plan, and your role.

get_reports

ChatGPT
Get weekly sentiment reports with AI summaries. Args: report_id: Specific report UUID for full detail (optional) limit: Number of recent reports (1-10, default: 5) Returns report summaries including worth_addressing, worth_addressing_status (one of "themes" | "clean" | "unavailable" | "legacy" — "unavailable" means the analysis failed or was skipped, so an empty worth_addressing list does not mean a clean week), TL;DR, highlights, lowlights, notable mentions, and week-over-week changes. An empty list may mean data collection is paused for this organization (trial or subscription ended) and older reports were already removed.

get_sentiment_overview

ChatGPT
Get aggregate sentiment analysis for a time window. Args: days: Lookback window in days (1-180, default: 7) Returns composite sentiment score, distribution (positive/neutral/negative), per-source breakdown, message count with week-over-week change.

get_sentiment_timeseries

ChatGPT
Get the daily sentiment trend for a time window. Args: days: Lookback window in days (1-180, default: 30) Returns one point per day — composite sentiment score, average message sentiment, and message count — ordered oldest to newest, plus a summary with the first and latest composite scores and the change between them. Days without generated insights are absent from the series rather than zero-filled.

get_stacks_ecosystem_sentiment

ChatGPT
This is the public Stacks ecosystem index across all opted-in projects — NOT this organization's own sentiment. Use get_sentiment_overview for this organization's own sentiment. Args: include_projects: Include the opted-in project leaderboard (scores, sparklines, volume bands). Defaults to false. Note this gates the leaderboard's DEPTH, not project identity: panel_joins (recently joined projects) and governance[].projects (the projects tracking each public vote) name projects in both variants, deliberately, because the public page renders them either way. Returns the same schema-versioned payload as the public HTTP index. This tool does not expose message text, authors, quotes, or exact per-project message counts.

search_messages

ChatGPT
Search community messages with filters. Args: source: Platform filter (discord, telegram, reddit, x, bluesky, lens, github, youtube, forum) start_date: Start date (ISO format YYYY-MM-DD) end_date: End date (ISO format YYYY-MM-DD) sentiment_min: Minimum sentiment score (1.0-10.0) sentiment_max: Maximum sentiment score (1.0-10.0) audience_tag: Filter by audience (community, support, engineering, trading) author: Author name search (partial match) keyword: Text search in message content relevance: Filter by relevance (on_topic or off_topic, default: on_topic) sort_by: Sort field (posted_at, sentiment_score, engagement) sort_order: Sort direction (asc or desc, default: desc) limit: Results per page (1-50, default: 25) offset: Pagination offset (0-10000) Returns messages with sentiment scores, audience tags, and metadata.

get_daily_insights

Claude

get_market_context

Claude

get_organization

Claude

get_reports

Claude

get_sentiment_overview

Claude

get_sentiment_timeseries

Claude

get_stacks_ecosystem_sentiment

Claude

search_messages

Claude

Example Prompts

Click any prompt to copy it.

App Stats

16

Tools

3

Prompts

ChatGPT, Claude

Platforms

Works with

ChatGPT
Claude

Data refreshed daily