How to score your next startup idea before you build it

Score every candidate on the same 5 axes before you touch code

You already have 2, 5, maybe 10 ideas sitting in a notes app. The question isn't where to find another one, it's which of the ones you already have is worth a month of your life. Browse live demand signals if you want to add to that shortlist, but the actual job here is ranking what's already on it.

Score each candidate on 5 axes, 0 to 3 points per axis, 15 points max: demand evidence, monetizability, competition and gap, build effort, and distribution reach. Add the 5 numbers. Build the one that clears 10 total with a demand score of at least 2. Kill anything that scores a 0 on competition, no matter how high the rest of the total runs. Everything in between goes on a 30 to 60 day watch list.

The rubric: what a 3 looks like versus a 1, per axis

Each axis has a hard definition per point value. Pull real numbers for each one before you score it, don't eyeball it.

Pull the demand numbers yourself, don't trust your gut

Run these before scoring the demand axis. Each one takes under a minute.

Hacker News, via the free Algolia search API, for any candidate topic:

curl -s "https://hn.algolia.com/api/v1/search?query=agent+budget+runaway+cost&tags=story&numericFilters=points%3E50" \
  | jq '.hits[] | {title, points, num_comments, url}'

GitHub, for existing tools already solving the same problem, sorted by stars:

curl -s "https://api.github.com/search/repositories?q=agent+budget+kill+switch+OR+mcp+credential+proxy&sort=stars&order=desc" \
  | jq '.items[] | {full_name, stargazers_count, pushed_at}'

Reddit and Product Hunt, as plain search operators, no API key needed:

site:reddit.com "agent" "budget" ("runaway" OR "burned through")
site:news.ycombinator.com "AI agent" ("budget" OR "cost overrun")
site:producthunt.com "budget" "AI agent"
"agent budget" pricing site:github.com

If a candidate returns fewer than 300 combined points, stars, and reactions across these 4 checks, it's a 1 on the demand axis. If it returns nothing at all, it's a 0. Don't score a 3 off vibes.

Pull the exact same evidence through Clawsmith's MCP

Clawsmith runs these searches on a schedule already, across 12 platforms, and keeps the totals. Instead of running 4 separate manual checks, one MCP call returns a signal's combined engagement as a single virality score:

Tool: search_signals
Args: {"query": "agent budget cost overrun", "limit": 5}

Response (real, pulled 2026-07-17):
{
  "slug": "ai-coding-agent-usage-cost-no-per-task-hard-ceiling",
  "signal_type": "issue",
  "virality_score": 102,
  "topic_tags": ["ai-agent-mcp", "agent-reliability"]
}

Pull the full brief on a candidate and the tool returns every source with its own engagement count, ready to sum by hand for the demand axis:

Tool: get_idea
Args: {"idea_slug": "agent-budget-kill-switch"}

issues_solved (6 sources, real):
- HN "Uber torches 2026 AI budget on Claude Code in four months": 402 points
- HN "Are the costs of AI agents also rising exponentially?": 306 points
- GitHub dipampaul17/AgentGuard: 168 stars
- HN "Show HN: AgentGuard": 73 points
- LeanOps benchmark (200 enterprise agent deployments): 42 reactions
- HN follow-on Uber thread: 28 points

6 sources across 3 platforms, 1,019 combined points and stars. That's a 3 on the demand axis under the rubric above.

Check who already built it before you score competition

The competition axis is the one people skip, and it's the one that kills the most ideas. Run the same GitHub search from above but read every result's README, not just the star count. A 160-star repo with 1 maintainer and no dashboard is a different competitive picture than a 1,900-star repo backed by a funded security company.

For SaaS competitors with pricing pages, search the exact feature plus "pricing" or "$/mo":

"credential proxy" agents pricing
"budget enforcement" AI agents "$" per month
site:producthunt.com "kill switch" agents

Cross-reference every result against the funding and revenue signals Clawsmith already tracks for a given idea:

Tool: get_competitors
Args: {"signal_id": "<signal-uuid-or-slug>"}

That call returns a table of every existing product, what it does, what's missing, and what it charges, so the gap you're about to build into is a documented one and not a hope.

Score 3 real ideas pulled from Clawsmith, side by side

Pulled these 3 via the MCP's get_ideas and get_idea tools, all inside Clawsmith's ai_agent_mcp and dev_tool_cli categories, all with real sourced evidence.

MCP credential proxy, 9 total. Demand is real, 24,008 secrets were found exposed in MCP config files according to GitGuardian's 2026 report, and the Infisical/agent-vault open source project alone sits at 1,900-plus GitHub stars. But 4 competitors already exist (Infisical, OneCLI, Peta, Alter), each with a named gap, no self-serve tier for solo developers, which caps the gap axis at 2 instead of 3. The build is 8 features deep (proxy, scoping, rotation, audit log, policy engine, team roles), so effort is a 1. 9 total, real signal, not urgent yet, revisit in 30 days.

Agent budget kill switch, 10 total. Uber burned through its entire 2026 AI budget on Claude Code in 4 months, a story that hit 402 points on Hacker News alone. The only existing tools are free and open source with 1 maintainer (AgentGuard, 160 stars) or unclear pricing (MintMCP), so the gap axis hits a full 3. Nobody's proven a direct paid comp in this exact niche yet, which caps money at 1. Total of 10, demand of 3, clears both thresholds. Build.

Search aggregation proxy, 6 total. Money looks incredible on paper, Tavily sold to Nebius for $275 million in February 2026 and Exa raised a $22 million Series A charging $5 per 1,000 queries. But an open source project (Search Stack MCP) already aggregates Tavily, Serper, and SearXNG behind one MCP tool, for free, and Tavily, Exa, and SerpAPI are already funded and selling into this exact use case. Competition scores a 0. Money being a 3 doesn't matter, a 0 on competition is an automatic kill.

The go/no-go rule, written as one function

Codify it so scoring stays consistent across every candidate you run through it:

def score(demand, money, gap, effort, dist):
    total = demand + money + gap + effort + dist
    if gap == 0:
        return total, "KILL (saturated, no gap)"
    if total >= 10 and demand >= 2:
        return total, "BUILD"
    return total, "WATCH (recheck in 30-60 days)"

score(3, 2, 2, 1, 1)  # MCP credential proxy -> (9, 'WATCH (recheck in 30-60 days)')
score(3, 1, 3, 1, 2)  # Agent budget kill switch -> (10, 'BUILD')
score(1, 3, 0, 2, 0)  # Search aggregation proxy -> (6, 'KILL (saturated, no gap)')

3 rules, in order: a 0 on competition kills it regardless of total. A total of 10 or more with demand at 2 or higher builds it. Everything else goes on the watch list and gets rescored in 30 to 60 days, because virality scores move. A signal sitting at 2,506 today can double by the next scan or get undercut by a new funded competitor.

Score every idea on your shortlist the same way, in the same sitting, so the numbers are comparable to each other and not just to your mood that day. If 2 ideas both clear 10, ship the one with the higher demand score first, since that's the axis that predicts whether anyone shows up on day 1. Check pricing once you've picked a winner and want the full feature prompts and lead list for the build, or keep browsing the startup ideas category for more candidates to add to the shortlist before you score the next round.