MCP trust layer
The trust and verification layer for MCP servers.
Federated directory across the Linux Foundation registry, Smithery, Glama, mcp.so, and the major community lists. Per-server trust scoring built on static source review, sandboxed runtime behavior, supply-chain provenance, and incident history. Scan any server before you install it.
Browse by purpose
Find what you need by what you're trying to do.
Directory
Trust-scored servers across every federated source. Best MCP servers, database servers, web search servers, free servers.
Security
The four spec-level attack patterns. Dated incidents. Best practices for builders. Scan any server before you install it.
Install configs
One-click setup for Claude Code, Claude Desktop, Cursor, Windsurf, ChatGPT, Continue, Cline, Zed.
Build a server
The three primitives. The Python SDK. A minimal working example in 30 lines.
Integrations
Bridge MCP to your workflow tools: n8n, Zapier, more on the way.
Enterprise
Org-wide trust scoring, allowlists, scheduled scans, internal verified-publisher tier, SSO.
Learn the protocol
Start from the published specification.
Reference content grounded in the LF AAIF spec, dated incidents, and the actual wire protocol. Read in the order that matches how you'll use it.
Hub · 23K/mo
Model Context Protocol
The full overview. Wire protocol, three primitives, three transports, governance, adoption.
Spoke
What is an MCP server
What a server actually does, mechanically. Lifecycle, primitives, privilege model.
Spoke
MCP architecture
JSON-RPC 2.0 wire mechanics, the initialize handshake, capability negotiation.
Compare
MCP vs API
REST and GraphQL vs MCP's standardized primitives for LLM context.
Compare
MCP vs RAG
Complementary by design. RAG for retrieval; MCP for live data and actions.
Compare
MCP vs A2A
Private tool-access (MCP) vs public agent-to-agent coordination (Google's A2A).
Origin
Anthropic MCP
Nov 2024 launch, Dec 2025 Linux Foundation donation, current governance.
Federation sources
10,970 MCP servers catalogued. 708 scanned. 48 scanned clean.
Federated across the Linux Foundation registry, the awesome-mcp-servers community list, Anthropic's reference set, MCPowered-curated entries, and user-submitted scans. Browse the scanned catalog or read the scoring methodology.
Why trust scoring
The MCP threat model is real and dated.
Sept 2025
postmark-mcp
First confirmed malicious MCP server in the wild. Silently BCC'd all outgoing emails to attacker.
Oct 2025
Smithery registry
Path-traversal vulnerability exposed a builder token granting root access to 3,000+ hosted apps.
Apr 2026
STDIO transport RCE
Core design flaw enabling authenticated RCE. Affected 150M+ downloads across Letta AI, LangFlow, Windsurf.
AgentSeal scanned 1,808 MCP servers in 2025. 66% had security findings.