Nine investor lenses, 25 live data sources, and a risk-asymmetric verdict that fires GO only when score ≥ 80, red flags = 0, and the cohort mirror holds. Reasoning on your tokens. Raw data on our pipes.
No card · BYOK supported (Perplexity, Tavily, Exa, Brave, DataForSEO, Firecrawl, Apify)
THE METHOD · LIVE
Every audit runs through market, pain, and competition lenses — live SERP, Reddit, funding rounds, hiring signals, 25 sources in total. Verdict lands in 90 seconds. No vibes.
See the methodOPERATOR-FIT · INTERACTIVE
Operator-fit lives between you, your market, and what you’re trying to build. Drag the surface. Watch the score deform in real time. That’s what an audit reads — and what a verdict locks in.
Drag to rotate · Spline · gated to capable devices
Every audit becomes a card on the channel — six witnesses, one verdict. GO. PIVOT. KILL. The wall never lies, and the ticker never stops.
Audit date: 2026-05-22 (refinalized with substrate v7) Verdict: GO · Zone: Star · Score: 91.0/100 (substrate) / 91.0/100 (crossed observer) Recommended next step: institutional seed ~$3M-$7M @ $20M-$40M pre-money, 18-24 month runway to Series A at $1M+ ARR
With CFPB Section 1033 already live for the largest institutions on April 1, 2026, mid-tier banks ($50B–$250B in assets) hit their phased deadlines in 2027–2028 while the rule sits under litigation and CFPB reconsideration (https://www.consumerfinance.gov/about-us/newsroom/cfpb-…
CMS's CY 2026 PFS final rule expanded DMHT payment to FDA-cleared/authorized ADHD devices, while declining GI and fibromyalgia coverage (https://www.sidley.com/en/insights/newsupdates/2025/11/cms-finalizes-key-medicare-reimbursement-policies-for-telehealth-chronic-disease-and-di…
Six failure modes are unique to agents — tool misuse, context loss, goal drift, retry loops, cascading multi-agent errors, silent quality degradation — none of which show up in HTTP-level metrics (https://latitude.so/blog/ai-agent-failure-detection-guide, https://coralogix.com/a…
Datadog's Feb 2026 telemetry showed 5% of LLM call spans erroring with 60% from rate limits; March 2026 dropped to 2% but rate limits stayed near a third (https://www.datadoghq.com/state-of-ai-engineering/, https://www.braintrust.dev/articles/agent-observability-complete-guide-2…
Agentic workloads are highly fragmented — GPUs face many small requests instead of long jobs, breaking the long-residency assumption of training-era schedulers (https://jimmysong.io/blog/gtc-2026-ai-native-infrastructure/, https://vexxhost.com/blog/gpu-capacity-crisis-ai-infrast…
CIRCIA's 72-hour reporting mandate goes live in May 2026 for critical-infrastructure-adjacent SMBs, while 88% of SMB breaches in 2025 involved ransomware (https://www.sentinelone.com/cybersecurity-101/cybersecurity/cyber-insurance-statistics/, https://www.security.org/insurance/…
28% of construction firms reported ICE-related workforce disruptions and ~10% lost workers directly, while the industry needs 349K net new hires in 2026 (https://www.constructionowners.com/news/construction-workforce-crisis-deepens-in-2026-amid-labor-shortages-and-ice-raids, htt…
Once Section 1033's regulated data-right kicks in, consumers gain pull-based access to 24 months of transactions across all covered banks (https://www.pymnts.com/bank-regulation/2026/data-aggregators-push-secure-access-as-rule-1033-rewrite-looms/, https://www.jdsupra.com/legalne…
CY 2026 PFS replaced the old 16-day RPM floor with codes for 2–15 days/month of monitoring plus 10-minute treatment-management billing (https://www.healtharc.io/blogs/how-cms-2026-medicare-proposed-rules-will-affect-telehealth-rpm/, https://www.orrick.com/en/Insights/2025/07/Med…
Audit date: 2026-05-22 (refinalized with substrate v7) Verdict: GO · Zone: Star · Score: 91.0/100 (substrate) / 91.0/100 (crossed observer) Recommended next step: institutional seed ~$3M-$7M @ $20M-$40M pre-money, 18-24 month runway to Series A at $1M+ ARR
With CFPB Section 1033 already live for the largest institutions on April 1, 2026, mid-tier banks ($50B–$250B in assets) hit their phased deadlines in 2027–2028 while the rule sits under litigation and CFPB reconsideration (https://www.consumerfinance.gov/about-us/newsroom/cfpb-…
CMS's CY 2026 PFS final rule expanded DMHT payment to FDA-cleared/authorized ADHD devices, while declining GI and fibromyalgia coverage (https://www.sidley.com/en/insights/newsupdates/2025/11/cms-finalizes-key-medicare-reimbursement-policies-for-telehealth-chronic-disease-and-di…
Six failure modes are unique to agents — tool misuse, context loss, goal drift, retry loops, cascading multi-agent errors, silent quality degradation — none of which show up in HTTP-level metrics (https://latitude.so/blog/ai-agent-failure-detection-guide, https://coralogix.com/a…
Datadog's Feb 2026 telemetry showed 5% of LLM call spans erroring with 60% from rate limits; March 2026 dropped to 2% but rate limits stayed near a third (https://www.datadoghq.com/state-of-ai-engineering/, https://www.braintrust.dev/articles/agent-observability-complete-guide-2…
Agentic workloads are highly fragmented — GPUs face many small requests instead of long jobs, breaking the long-residency assumption of training-era schedulers (https://jimmysong.io/blog/gtc-2026-ai-native-infrastructure/, https://vexxhost.com/blog/gpu-capacity-crisis-ai-infrast…
CIRCIA's 72-hour reporting mandate goes live in May 2026 for critical-infrastructure-adjacent SMBs, while 88% of SMB breaches in 2025 involved ransomware (https://www.sentinelone.com/cybersecurity-101/cybersecurity/cyber-insurance-statistics/, https://www.security.org/insurance/…
28% of construction firms reported ICE-related workforce disruptions and ~10% lost workers directly, while the industry needs 349K net new hires in 2026 (https://www.constructionowners.com/news/construction-workforce-crisis-deepens-in-2026-amid-labor-shortages-and-ice-raids, htt…
Once Section 1033's regulated data-right kicks in, consumers gain pull-based access to 24 months of transactions across all covered banks (https://www.pymnts.com/bank-regulation/2026/data-aggregators-push-secure-access-as-rule-1033-rewrite-looms/, https://www.jdsupra.com/legalne…
CY 2026 PFS replaced the old 16-day RPM floor with codes for 2–15 days/month of monitoring plus 10-minute treatment-management billing (https://www.healtharc.io/blogs/how-cms-2026-medicare-proposed-rules-will-affect-telehealth-rpm/, https://www.orrick.com/en/Insights/2025/07/Med…
We provide raw data, deterministic math, and persistence. Anthropic charges you for reasoning. We charge for storage and pooled data — or nothing if you BYOK.
The 15 markdown skills install into ~/.claude/skills. Claude follows them step-by-step. Edit them. Fork them.
42 tools across 6 categories. SERP, Trends, app stores, Reddit, web search, pure-math signals — Claude orchestrates.
Audit history with full markdown reports, lens-by-lens breakdown, quota meters, investor uploads, billing — all yours.
The audit-idea master skill orchestrates 9 sub-skills + dealbreakers synthesis. Each lens runs against real data, not vibes.
Live keyword volumes via DataForSEO Trends + Keyword Overview. SERP-classified competitors with category buckets and gap analysis. Reddit pain density via site-scoped SERP counting.
Founder track record web-grounded. Github stars, hiring posts, funding rounds — counted, not guessed.
3 perspectives: operator, consumer, trends. 10 business-model candidates.
Search velocity, urgency composite, build complexity.
Weighted across 6 pitch dimensions (team 25, P/S 20, traction 20, comp 15, GTM 10, fin 5). Star / Green / Yellow / Orange / Red — with prioritised remediation roadmap.
1-pager teaser, blurb (5 hook variants), email template (5 subject lines), best-fit investor matches with 0–100 score per investor.
Sortable by zone, searchable by title, grouped by month. Audits land here automatically the moment Claude finalises a report.
Plug your own keys to skip pooled metering, unlock subscription discounts, and route through accounts you already pay for.
Most audit tools sell you reasoning. We sell you the rails — your reasoning, your model, your data, your reports.
Server has zero LLM API keys. We can't see prompts, can't peek at audits, can't read your inputs.
AES-256-GCM envelope encryption with KMS-rotated master key. We see plaintext once on save, then never again.
All 15 skills are markdown. Read them. Edit them. Fork them. Audit-grade transparency on every step.
Plain-English explainers on observer profiles, kill criteria, MCP, BYOK economics, and the methodology behind every verdict.
MethodologyMost dead startups are dead three months before the founder admits it. Pre-committed kill criteria — three quantitative shutdown triggers, set before week one — are the only mechanism that converts 'maybe one more pivot' into a binary verdict. Here is the rubric, the failure modes, and the numbers.
ProductAn audit verdict is a snapshot. The Founder Loop turns it into a 12-week build cycle with kill criteria, weekly experiments, an evidence log, a cohort mirror, a pivot detector, and a channel playbook — each one a deterministic surface, not a vibe.
MethodologyChatGPT, Gemini, and Claude will hand you a hundred 'startup ideas' in a minute. Almost none survive the first SERP check. Here is why generic LLMs fail at idea quality, and what grounding fixes.
All plans include the full 42-tool MCP and 15 skills. BYOK calls don't count against quota.
FAQ
The short, no-fluff version. For depth see the docs. the docs.
ideaudit is an open-source Model Context Protocol server plus 15 markdown skills that turn Claude Desktop, Claude Code, or any Anthropic SDK app into an investment-grade startup idea-audit pipeline. Reasoning runs on your Anthropic tokens; raw data fetch (SERP, Trends, Reddit, review-sites) and audit persistence run on the ideaudit MCP server. You get a stage-aware verdict, a 0–100 score across six lenses, and a saved audit in your dashboard.
Three steps. (1) Install the 15 skills locally with curl https://audit.inite.ai/install.sh | sh. (2) Generate an API key at audit.inite.ai/settings/keys — the plaintext is shown once. (3) Add the MCP server to your Claude config. Then in Claude type "audit this idea: …" and the report streams into your dashboard. The whole loop is two minutes.
Yes. Claude reasoning runs on your tokens, so Claude Desktop, Claude Code, or your SDK app needs an active Anthropic account. ideaudit never spends tokens on your behalf — it only provides the deterministic data, math, and storage layer. This keeps the cost split honest: you pay the LLM provider for thinking, you pay us for data and infra.
Google SERP, Google Trends, Reddit, Product Hunt, Trustpilot, G2, GitHub, Crunchbase signals, and review-site aggregations. Bring-your-own-key is supported for Perplexity, Tavily, Exa, Brave, DataForSEO, Firecrawl, and Apify — useful if you prefer your own data quotas, want lower-latency reads, or already pay for a research stack you trust.
Yes — three flavours, one endpoint. Claude Desktop reads MCP config from the JSON config file. Claude Code CLI registers it via `claude mcp add --transport sse ideaudit https://audit-api.inite.ai/mcp --header "Authorization: Bearer …"`. The Anthropic Messages API takes an `mcp_servers` array with the same URL and bearer token. All three share the same audit history.
The 15 skills are plain markdown in github.com/inite-ai/initeaudit/skills — read them, edit them, fork them, drop them into your own /.claude/skills/. The MCP tools have documented JSON schemas. Audit-grade transparency on every step Claude follows. The hosted MCP server itself is closed source for now (it carries the data-provider quotas), but everything that shapes Claude's reasoning is on disk in your home directory.
A stage-aware verdict — GO, REFINE, or KILL — with a 0–100 investor-readiness score. Six lenses get scored independently: Team, Problem-Solution, Traction, Competition, GTM, Financials. Each carries a confidence and a red-flag flag. A cross-lens consensus check surfaces contradictions (e.g. "market: 30% CAGR" vs "competition: saturated, FAANG-dominated") and those feed into the final dealbreakers pass before the verdict is locked.
