let agents kill
bad ideas
at scale
One REST or MCP call turns any idea into a GO / CONDITIONAL_GO / PIVOT / NO_GO verdict with scores and live-web evidence. $0.20 a call. And we never keep your idea — it is deleted right after the verdict is delivered.
one endpoint.
any agent.
Claude Code, Cursor, or your own agent loop — same key, same prices, same balance over REST and MCP.
# 1. Get a key — no email, no sign-up. Shown once.
curl -X POST https://api.ideakiller.app/agent/v1/keys
# 2. Validate an idea (light tier, $0.20, ~90 s).
curl -X POST "https://api.ideakiller.app/agent/v1/validations?wait=60" \
-H "X-API-Key: ika_..." -H "Content-Type: application/json" \
-d '{"idea_name": "AI tax filing for freelancers",
"description": "Automates expense categorization, quarterly estimates and annual filing for US freelancers.",
"price_point": 15, "business_model": "subscription",
"tier": "light"}'
# 3. If the answer is 202, poll the poll_url.
curl "https://api.ideakiller.app/agent/v1/validations/{call_id}?wait=60" \
-H "X-API-Key: ika_..."// claude_desktop_config.json, cursor mcp.json, any MCP client
{
"mcpServers": {
"ideakiller": {
"url": "https://api.ideakiller.app/mcp",
"headers": { "Authorization": "Bearer ika_..." }
}
}
}
// Claude Code one-liner:
// claude mcp add --transport http ideakiller \
// https://api.ideakiller.app/mcp \
// --header "Authorization: Bearer ika_..."
// Tools your agent gets:
// validate_idea · get_validation · get_balance · create_topupimport requests, time
API = "https://api.ideakiller.app/agent/v1"
H = {"X-API-Key": "ika_..."}
r = requests.post(f"{API}/validations?wait=60", headers=H, json={
"idea_name": "AI tax filing for freelancers",
"description": "Automates expense categorization, quarterly estimates and annual filing for US freelancers.",
"price_point": 15,
"business_model": "subscription",
"tier": "light",
}).json()
while r["status"] == "RUNNING":
time.sleep(15)
r = requests.get(r["poll_url"], headers=H).json()
print(r["result"]["verdict"]["type"]) # GO / CONDITIONAL_GO / PIVOT / NO_GO
print(r["result"]["scores"]["overall"]) # 0-100const API = "https://api.ideakiller.app/agent/v1";
const headers = {
"X-API-Key": "ika_...",
"Content-Type": "application/json",
};
let res = await fetch(`${API}/validations?wait=60`, {
method: "POST",
headers,
body: JSON.stringify({
idea_name: "AI tax filing for freelancers",
description: "Automates expense categorization, quarterly estimates and annual filing for US freelancers.",
price_point: 15,
business_model: "subscription",
tier: "light",
}),
}).then((r) => r.json());
while (res.status === "RUNNING") {
await new Promise((r) => setTimeout(r, 15000));
res = await fetch(res.poll_url, { headers }).then((r) => r.json());
}
console.log(res.result.verdict.type); // "CONDITIONAL_GO"
console.log(res.result.scores.overall); // 63key → top-up →
validate → poll.
No dashboards, no OAuth dance. Four calls and your agent has a verdict with receipts.
key
/agent/v1/keys
One POST. No email, no sign-up, no dashboard. The key is shown once — your agent stores it.
top-up
/agent/v1/topups
Prepaid balance, minimum $20. The API returns a checkout_url — hand it to a human once, then run autonomously.
validate
/agent/v1/validations
POST the idea with ?wait=60. A light call lands in about 90 seconds and costs $0.20.
poll
/agent/v1/validations/{call_id}
Still running? You get a 202 and a poll_url. The result is delivered once, then deleted from our side.
what your agent
gets back.
Structured JSON, not prose to parse. Verdict, conviction, six scores and evidence with live-web sources — ready for the next step in the agent loop.
{ "status": "COMPLETED", ← 202 + poll_url while still running "tier": "light", ← light $0.20 · full $0.50 "cost_usd": 0.20, "result": { "verdict": { "type": "CONDITIONAL_GO", ← GO · CONDITIONAL_GO · PIVOT · NO_GO — branch on it "conviction": 55, ← 0–100 confidence in the verdict "conditions_for_success": […], "kill_criteria": [{ "trigger": …, "action": …, "deadline": … }] ← re-checks your agent can schedule }, "scores": { ← all 0–100 "demand": 74, "competition": 42, ← competition: higher = more room "problem_fit": 72, "economics": 58, "risk": 62, "overall": 63 ← risk: higher = more danger }, "demand": {…}, ← market size, search volume, sources "competitors": {…}, ← competitors, gaps, sources "unit_economics": {…}, ← LTV, CAC, churn, payback "risks": {…}, ← failure modes with severity + mitigation "pivot": null ← pivot ideas when the verdict is PIVOT (full tier) } } agents that check
before they build.
startup-scout agents
Your scout surfaces a hundred raw ideas a night. Make it prove them: only ideas with a verdict survive to your morning inbox.
vc deal-flow screening
Screen inbound decks at 3am. Verdict, conviction and six scores attached before a partner ever opens the PDF.
self-filtering generators
Generate → validate → kill. Your idea generator eats its own dog food and only shows users the ideas that pass.
founder copilots
Before your copilot lets a founder burn the runway on a pivot, it checks the pivot. Two minutes of humility, on demand.
it passed.
now name it.
A GO verdict is worth nothing if the name hides what you sell. The domain finder builds names on words your buyers really search (real Bing volumes), checks them in the registry and returns 10 free ones with reasons. $0.20 a search — same key, same balance.
cheaper than
the prompt.
Your agent burns more than $0.20 thinking about whether to build. Pay per verdict, only when it is delivered — and the idea is deleted right after.
- verdict + conviction
- 5 scores + overall
- live-web evidence per score
- default tier
- everything in light
- deeper web research
- competitor + risk detail
- pivot ideas on PIVOT
- up to 10 free domains
- built on real Bing search volumes
- niche + competitor research
- a reason for every name
prefer a form?
same engine, no code.
The web app runs the same analysis behind a three-minute flow. Free credits every month, no API key needed.
we never keep
your ideas.
Other tools feed your idea into a training set or a trend report. Here, the idea and the report are deleted right after the verdict is delivered. Gone. Not "anonymized" — deleted.
questions,
answered.
Does my agent need an account to call the API?
How does an agent pay for calls?
What happens to my idea after the verdict?
What if my agent sends garbage or an empty idea?
What if a validation call fails midway?
Is there an MCP server?
Where are the machine-readable docs?
What are the rate limits?
stop funding
the wrong thing.
One call. Twenty cents. A verdict your agent can act on — and an idea we never keep.