Here is your prompt.
Your first 50.

We're going to find my first 50 users with a four-phase method. At the end of each phase, show me the result and wait for my approval before moving on. Don't send anything: I send every message myself.

PHASE 1 · THE PAIN IN MY USER'S WORDS
Ask me what my app does, who it's for, 2 or 3 competitors, the language and the country.
Then propose 15 search phrases the way my user would write them, in three groups:
- Complaint: "I'm so tired of…", "my inventory never adds up…".
- Question: "does anyone know an app for…?", "what do you use for…?".
- Competitor: "[competitor] doesn't work for…", "alternative to [competitor]".
For X, turn them into queries with operators: lang:[language] -is:retweet and, if relevant, words from the country.
Stop. Wait for me to pick the phrases.

PHASE 2 · SEARCH WHERE THEY'RE ALREADY COMPLAINING
Use only the tools I have connected; if one is missing, tell me how to connect it and continue with the rest:
- X: the X MCP in read-only mode, or GET /2/tweets/search/recent with X_BEARER_TOKEN (last 7 days).
- YouTube: the vidIQ MCP (videos that took off in my niche and their comments) or the YouTube Data API with YOUTUBE_API_KEY.
- Optional: Exa (EXA_API_KEY) for forums and communities; 1 to 3 star reviews of my competitors only to understand what they're missing, never to contact anyone.
You can split each source into its own parallel task. Before searching, tell me how many queries you'll run and the estimated cost (posts read × US$0.005 on X, vidIQ credits, Exa searches). Wait for my approval.
Keys live in .env (and .env in .gitignore). Never ask me for them in the chat or write them in the code.

PHASE 3 · QUALIFY BEFORE WRITING
- Open every link and check the exact quote. If you can't verify it, discard it. Never invent a quote.
- Each person counts once, with their strongest quote.
- Score from 0 to 10: complains about the pain +3, asks for a recommendation +3, mentions a competitor +2, urgency +2, older than 30 days −2.
- Temperature: hot (8 to 10), warm (5 to 7), cold (under 5, observe only).
Deliver the top 50 by score. Stop.

PHASE 4 · THE FIRST MESSAGE
For each person, write a 2 or 3 sentence message that quotes what they said, with no link, doesn't sell and ends with a question.
Save everything to first-50.csv with: #, user, source, link, exact_quote, date, signals, score, temperature, why_it_fits, first_message, status (pending), replied (empty).
Show me the table sorted from hot to cold and explain how to measure: how many I sent, how many replied, how many agreed to try the app.

RULES THAT DON'T BREAK
1. No automated sending: don't use direct message, posting or email tools, even if the MCP includes them.
2. Public data only. Don't look for emails or phone numbers and don't cross-match a person across platforms.
3. Respect each API's terms and limits. No forbidden scraping and no getting around rate limits.
4. If someone says no or doesn't reply, one follow-up at most.