Find Content Ideas from Questions People Already Ask

Made by Desearch (desearch.ai).

This content ideas finder is for marketers and founders who want post ideas from questions people already ask. Paste one saved prompt into Claude or Cursor with the Desearch MCP server connected, and you get up to 10 questions, a post idea for each and a source link. A small Python script prints the same kind of table.

The problem

A content plan often starts from a blank page. Searching Reddit, Hacker News and X by hand means opening a lot of threads and trying to remember which questions came up. What helps is one table: the question, a post idea, a format and the link, so anyone can check the source before writing.

What the app does

• Takes a niche in plain words, such as "starting a newsletter".
• Searches Reddit with the Desearch ai-search tool for "questions beginners ask about" that niche.
• Runs the same search on Hacker News, unless you pass --no-hackernews.
• Searches public X posts with the x-search tool (20 posts by default).
• Pulls question sentences that mention the niche, keeps one per thread or post, and suggests a format: carousel for lists and tips, short video for how-tos, thread for the rest.
• Prints a table of up to 10 questions with a post idea, the format, the source link and the date when one is shown.

How it works

1. Connect once. Add the hosted Desearch MCP server (https://mcp.desearch.ai/mcp) to Claude or Cursor, with your Desearch API key in the Authorization header.
2. Search. The saved prompt calls ai-search with Reddit, then Hacker News for a tech niche, then x-search with count 20.
3. Plan. Your assistant lists up to 10 questions from those results only, with a post idea, a hook, a format and the source link. It does not name individual posters.

The code

No code is needed for the saved prompt. For people who want a script, these are the search arguments from content_ideas.py (lines 45 to 56). The script reads the API key from a key file (~/.desearch-api-key, or --key-file PATH) and prints only the last 4 characters of its sha256 fingerprint.

def build_ai_args(niche, source):
    """ai-search arguments for one source (reddit or hackernews). No domain filters."""
    if source not in ("reddit", "hackernews"):
        raise ValueError("source must be reddit or hackernews")
    return {"prompt": f"questions beginners ask about {clean_niche(niche)}",
            "tools": [source], "result_type": "LINKS_WITH_FINAL_SUMMARY"}


def build_x_args(niche, count=20):
    """x-search arguments. OR terms stay inside parentheses."""
    q = f'({clean_niche(niche)}) ("how do i" OR "how do you" OR "any tips") lang:en -is:retweet'
    return {"query": q, "count": max(1, min(int(count), 100))}

The MCP call, from the same file (lines 222 to 236):

async def call_tools(calls, key):
    """Run [(tool, args), ...] in one MCP session; return parsed results."""
    from mcp import ClientSession
    from mcp.client.streamable_http import streamable_http_client
    from mcp.shared._httpx_utils import create_mcp_http_client

    out = []
    async with create_mcp_http_client(headers={"Authorization": key}) as hc, \
            streamable_http_client(MCP_URL, http_client=hc) as streams:
        async with ClientSession(streams[0], streams[1]) as s:
            await s.initialize()
            for tool, args in calls:
                res = await s.call_tool(tool, args, read_timeout_seconds=180)
                out.append(parse_content([c.model_dump() for c in res.content]))
    return out

The folder has saved-prompt.md (the no-code version), content_ideas.py, README.md, sample_output.md, sample_plan.md and offline tests in tests/.

View the code

Endpoints

• MCP tool ai-search (tools: reddit, hackernews)
• MCP tool x-search

Example output

Saved sample from a run on 2026-10-17 for the niche "starting a newsletter". This page shows that stored request and response. It does not run a live demo.

Sample request

build_ai_args() and build_x_args() send these arguments. The script runs the Reddit search, the Hacker News search, then the X search.

{
  "prompt": "questions beginners ask about starting a newsletter",
  "tools": ["reddit"],
  "result_type": "LINKS_WITH_FINAL_SUMMARY"
}

{
  "prompt": "questions beginners ask about starting a newsletter",
  "tools": ["hackernews"],
  "result_type": "LINKS_WITH_FINAL_SUMMARY"
}

{
  "query": "(starting a newsletter) (\"how do i\" OR \"how do you\" OR \"any tips\") lang:en -is:retweet",
  "count": 20
}

Sample response

The script printed this, from sample_output.md:

Niche: starting a newsletter
Read 40 threads and posts Desearch returned; found 10 questions.

The saved prompt, tested 2026-10-17 09:26 Asia/Tbilisi, returned 10 Reddit threads, 10 Hacker News threads and 20 X posts. The first three rows of the script table:

The sample ends with this line:

ai-search cost: $0.0080

Reddit and Hacker News results come without dates, and X results mix newer and older posts. Check the date column, or open the link, before you build a post on a question.

Run it yourself

No code: connect the Desearch MCP server in Claude or Cursor, paste saved-prompt.md, and replace the niche.

Script:

git clone https://github.com/Desearch-ai/desearch-py-examples && cd desearch-py-examples/content-ideas-finder-mcp
pip install -r requirements.txt
printf '%s' 'YOUR_KEY' > ~/.desearch-api-key && chmod 600 ~/.desearch-api-key   # key from console.desearch.ai
python content_ideas.py "starting a newsletter"

Useful flags: --no-hackernews (non-tech niches), --no-x, --count (X posts to read, default 20), --limit (questions to keep, default 10), --raw (save the responses), --from-raw (rebuild the table from a saved file, with no API calls).

What it costs

Each ai-search call cost $0.004 on 2026-10-17. The sample prints ai-search cost: $0.0080 for its two ai-search calls. See pricing.

Ideas to extend it

• Run the same niche once a month and compare the questions.
• Skip Hacker News for a non-tech niche with --no-hackernews.
• Use the saved prompt when you want hooks written from the threads, and the script when you want the table from the same searches.