📊 MCP Ecosystem Dashboard

Live statistics from MarketNow's audit of 8,764 MCP servers. Updated weekly.

8,764Total MCP Servers

5,120Sentinel Audited

206L2.5 Sandbox Tested

65Free Skills

Security Score Distribution

10/10 — 69 servers (clean)

8/10 — 10 servers

7/10 — 9 servers

5/10 — 9 servers (medium risk)

2/10 — 6 servers (high risk)

0/10 — 103 servers (failed to start)

Based on 206 L2.5 gVisor sandbox audits. The 103 "0/10" servers failed to start under gVisor — likely due to incompatible syscalls, not necessarily malicious.

Top Categories

Category	Servers	Share
Developer Tools	6,137	70.0%
AI/ML	433	4.9%
Data	411	4.7%
Search	333	3.8%
Finance	283	3.2%
Communication	288	3.3%
Productivity	328	3.7%
Security	224	2.6%
Other	327	3.7%

Sentinel Audit Pipeline

Layer	Status	Coverage	What it does
L1.5	✅ Live	5,120 skills	Static analysis (deps, secrets, licenses)
L1.6	✅ Live	5,120 skills	Pattern-based behavioral analysis
L2 v2.0	✅ Live	206 skills	Active probe (60+ adversarial inputs)
L2.5	✅ Live	206 skills	gVisor sandbox (userspace kernel)
L3	⏳ Q1 2027	—	Firecracker microVM (KVM isolation)
L4	⏳ Q4 2026	—	Supply chain attestation (SLSA)
L5	⏳ Q3 2027	—	Third-party audit (Trail of Bits)

Key Findings

3 servers removed for leaking environment variables via tools/call (args passed to eval() without sanitization)
12 servers had hardcoded API keys in source code
47% of servers make network calls (potential exfiltration risk)
1 server attempted ptrace() — blocked by gVisor
1 server attempted bpf() — blocked by gVisor
0 container escapes detected under gVisor sandbox

Adversarial Input Categories Tested

Path Traversal (critical) SSRF (critical) Command Injection (critical) Credential Access (critical) SQL Injection (high) Prompt Injection (high)

60+ adversarial inputs sent via real MCP protocol (initialize, tools/list, tools/call)

npm Package Stats

Loading npm stats...