All data validated against primary sources Feb 25, 2026. Confidence: HIGH (primary source verified), MEDIUM (verified with caveats), LOW (unverifiable).
| Metric | Value | Source | Confidence |
|---|
| AI agents market 2025 | $7.6B (GVR) / $8.0B (FBI) | Grand View Research, Fortune Business Insights | HIGH |
| AI agents market 2030 | $50.3B (GVR) / $52.6B (M&M) / $48.3B (BCC) | Three independent firms converge | HIGH |
| AI agents market 2033 | $183B | Grand View Research (49.6% CAGR) | HIGH |
| AI agents market 2034 | $251B | Fortune Business Insights (46.6% CAGR) | HIGH |
| Agent builder platforms 2026 | $5.0B | Gartner | HIGH |
| Agent builder platforms 2029 | $13.7B | Gartner | HIGH |
| Agentic AI spending 2029 | $1.3T (26% of IT) | IDC | HIGH |
| Coding/dev tools AI CAGR | 52.4% | MarketsandMarkets | HIGH |
| Registry | Count | Curation Level |
|---|
| Official MCP Registry | 518 | Minimal (canonical metadata feed) |
| Smithery | 7,300+ | Moderate (registry + hosting) |
| PulseMCP | 8,610+ | Curated (omits low-quality) |
| Glama.ai | 17,697 | Comprehensive (quality scoring) |
| MCP.so | 17,867 | Low (community-driven) |
| GitHub MCP repos | ~20,000 | None (raw search results) |
| Registry | Count | Notes |
|---|
| SkillsMP | ~96,751 | GitHub auto-aggregator; previously claimed 270K |
| skills.sh (Vercel) | 60,000+ | Launched Jan 2026; 110K installs in 4 days; Snyk partnership for security scanning |
| ClawHub | 5,705+ (expanded: 10,700+) | 1.5M+ downloads; 341→1,184 malicious skills purged (ClawHavoc); 12% malicious at peak |
| SkillHub | 7,000+ | AI quality scoring (S/A/B/C grades) |
Independent analysis of 4,784 skills across 5 registries (pradeep.md, Feb 2026):
- 28% are duplicates (only 2,990 unique names)
- 12% are empty implementations
- 59% ship executable scripts
- Only 2% declare sandboxing
- Only 31% score 90+ quality rating
Defensible unique skill count: ~130-150K (after dedup and quality filtering)
| Metric | Value | Source | Confidence |
|---|
| MCP monthly SDK downloads | 97M+ | Pento / Anthropic | MEDIUM |
| MCP donated to Linux Foundation | Dec 9, 2025 | Anthropic (co-founded AAIF with Block, OpenAI) | HIGH |
| AAIF backing | Google, Microsoft, AWS, Cloudflare, Bloomberg | Multiple sources | HIGH |
| AAIF scope | Unifying MCP, AGENTS.md, goose protocols | Linux Foundation AAIF charter | HIGH |
| Next MCP spec revision | Expected June 2026 (expanded long-running workflows, Nov 2025) | MCP roadmap | MEDIUM |
| SKILL.md platform adoption | 26+ platforms | Multiple | HIGH |
| Official MCP Registry approach | ”Canonical feed” — deliberately NOT a marketplace | MCP blog, Gentoro analysis | HIGH |
Critical context: Anthropic deliberately designed the MCP Registry as minimal metadata infrastructure, NOT a user-facing marketplace. They explicitly leave discovery, search, and commerce to third parties. This is bullish for Findable.
| Metric | Value | Source |
|---|
| GitHub stars | ~215K+ (fastest to 100K in history) | GitHub |
| Active instances | 100K+ (42K+ publicly exposed) | Censys, Bitsight |
| ClawHub skills | 5,705+ (expanded registry: 10,700+) | ClawHub |
| Skill downloads | 1.5M+ | ClawHub |
| Creator | Peter Steinberger (joined OpenAI Feb 14, 2026) | TechCrunch |
| Project status | Moving to independent open-source foundation | Steinberger blog |
| Metric | Value | Source |
|---|
| Registered agents | 2.66M | MoltBook / Wikipedia |
| Real human owners | ~17,000 (88 agents per person avg) | Wiz security audit |
| Human observers (first week) | 1M+ | NBC News |
| Security breach | Unsecured DB, 1.5M API keys + 35K emails exposed | 404 Media |
| Revenue | $0 (only a speculative $MOLT memecoin) | Multiple |
- ClawHub’s security crisis validates Findable Shield: ClawHavoc found 1,184 malicious skills (12% of registry). 41.7% had serious vulnerabilities. This is exactly the problem our security scanner solves.
- ClawHub = addressable market: 5,705+ skills with 1.5M+ downloads. If Findable can index and trust-score ClawHub skills, we access the largest agent skill ecosystem.
- MoltBook’s catastrophic security (unsecured database, prompt injection, API key exposure) is the poster child for why agent infrastructure needs a trust layer.
- No monetization despite massive demand: 2.66M agents, $0 revenue. The commerce opportunity is wide open.
- OpenClaw skill format = SKILL.md: OpenClaw skills use the same SKILL.md format Findable indexes. Cross-platform coverage is natural.
| Study | Sample | Finding | Source |
|---|
| Enkrypt AI | 1,000 servers | 32% have critical vulns, avg 5.2 per server, 0% have security docs | Enkrypt AI report |
| earezki.com | 518 (official registry) | 41% lack authentication | earezki.com audit (Feb 2026) |
| BlueRock/Microsoft | 7,000+ servers | 36.7% have latent SSRF vulnerability | BlueRock analysis |
| Backslash Security | 15,000+ servers | ~50% dangerously misconfigured | Backslash blog |
| Snyk | 3,984 ClawHub skills | 7.1% leak credentials (API keys, CC numbers, PII) | Snyk research (Feb 2026) |
| VirusTotal | 17,845 GitHub repos | 8% flagged malicious | VirusTotal blog (Jun 2025) |
| Incident | Date | Impact |
|---|
| ClawHavoc: 1,184 malicious skills on ClawHub | Feb 2026 | 12% of registry; Atomic macOS Stealer (AMOS), credential theft, crypto drainers |
| MoltBook database breach | Jan 31, 2026 | 1.5M API keys + 35K emails exposed via unsecured database |
| Summer Yue email deletion | Feb 23, 2026 | OpenClaw agent deleted 200+ emails from Meta AI researcher’s inbox |
| Cline CLI supply chain attack | Feb 17, 2026 | Compromised npm package secretly installed OpenClaw on ~4,000 developer machines |
| CVE-2025-6514: mcp-remote RCE | Jun 2025 | 437K+ installations affected |
| Anthropic Git/Filesystem MCP server RCE | Jan 2026 | Anthropic’s own servers exploitable |
| Supabase Cursor agent data exfiltration | Mid-2025 | Customer data breach |
| Asana MCP privacy breach | Jun 2025 | Customer data bleed |
| GitHub MCP vulnerability: repo issues hijack agents | May 2025 | Supply chain attack vector |
| Standard | Body | Date |
|---|
| OWASP Top 10 for Agentic Applications | OWASP GenAI Security Project | Dec 2025 |
| OWASP MCP Top 10 | OWASP Foundation | 2026 |
| CoSAI MCP Security Whitepaper | Coalition for Secure AI | Jan 2026 |
| CoSAI Secure-by-Design Agentic Systems | CoSAI | 2026 |
| NIST Agent Identity/Authorization | NIST/NCCoE | Feb 2026 |
| Cisco AI Defense for Agentic AI | Cisco | Feb 2026 |
| Platform | Model | Revenue | Status |
|---|
| 21st.dev | Freemium ($16-32/mo) | ~$400/mo MRR | Only documented individual success |
| Apify MCP | Pay-per-event (80% rev share) | Devs earn up to $2K/mo | Active, real transactions |
| Composio | Enterprise managed MCP | $2M ARR, 200+ customers | Active (infra, not marketplace) |
| Ref (ref.tools) | Credits ($9/1K) | Unknown | Active |
| ClawMarket | Crypto escrow (USDT) | 40 listings, minimal | Niche |
| MCP Hive | Pay-per-request | Pre-launch Mar 2026 | Not yet live |
| MCPize | Usage-based | Unknown | Active |
Total ecosystem paid-skill revenue: <$100K/month. The agent skill commerce market is pre-revenue.
| Provider | What | Status |
|---|
| Apify | Full hosting + billing + distribution | Active, most mature |
| Nevermined | Agent billing (MCP, A2A, x402, fiat + crypto) | Active |
| Moesif | API analytics + MCP metering | Adding MCP support |
| Stripe | Standard payment APIs | Not MCP-specific |
| Signal | Source | Confidence |
|---|
| 57% of orgs have agents in production | LangChain State of Agent Engineering | MEDIUM |
| 33% of enterprises run agentic AI in production | Battery Ventures survey | MEDIUM |
| 48% plan agent deployment within 12 months | Battery Ventures survey | MEDIUM |
| 79% say AI agents being adopted | PwC survey | MEDIUM |
| 40% of agentic AI projects to be canceled by 2027 | Gartner | HIGH |
| 67% will maintain AI spending even in recession | KPMG AI Pulse Survey | MEDIUM |
| Half plan $10-50M for agentic architectures | KPMG | MEDIUM |
| What We Don’t Know | Why It Matters |
|---|
| ”Agent skill marketplace” TAM from any analyst | Our TAM is internally estimated, not analyst-sourced |
| Active MCP server users (not downloads) | 97M downloads could be 1M or 10M active users |
| Developer willingness to pay for registry features | No survey data on this specific question |
| Agent-autonomous tool discovery adoption | Still experimental; not mainstream |
| Findable-specific demand validation | No pre-launch survey or waitlist conversion data |