On May 12, 2026, Manifold Security expanded its Manifest platform to index and automatically score 7,700+ MCP (Model Context Protocol) servers — every server currently listed in the official MCP Registry, which now tracks over 206,000 total assets. The expansion makes Manifest the most comprehensive automated security scoring system for the MCP ecosystem.
The tool is freely accessible at manifest.manifold.security.
Why MCP Server Security Matters Now
The MCP ecosystem has grown explosively since Anthropic introduced the protocol. It’s now the dominant standard for giving AI agents access to external tools, data sources, APIs, and services. When an agent needs to read a file, query a database, or call an external service, it typically does so through an MCP server.
That makes MCP servers part of the AI agent supply chain — and supply chain security is only as strong as the weakest component. An MCP server that contains prompt injection vulnerabilities, coercive instructions, or comes from an unverified publisher creates attack surface for every agent that connects to it.
Until tools like Manifest existed, vetting an MCP server required manual code review — feasible for a handful of servers, impossible for organizations connecting to dozens.
How Manifest Scoring Works
Each MCP server in the Manifest index receives a composite Manifest Score built from two components:
1. Lineage Score — Publisher Trust This measures the trustworthiness of who built the server:
- Publisher verification status
- Repository age (how long the codebase has existed)
- Commit patterns (consistent maintenance activity vs. abandoned or burst-published repos)
- Publisher track record across other published assets
2. Safety Score — Content Risk This assesses what the server actually does and how it’s implemented:
- Prompt injection risk indicators (instructions embedded in server responses that could redirect agent behavior)
- Coercive instructions (instructions that attempt to override agent guidelines)
- Interface contradictions (mismatches between declared capabilities and actual behavior)
The two scores combine into the composite Manifest Score, which gives security teams a single signal to prioritize review — though Manifold recommends examining component scores for any server in a production stack.
From Free Tool to Enterprise Runtime
The free tier at manifest.manifold.security provides on-demand scoring and browsing of the full 7,700+ server index. This is the right starting point for any organization currently evaluating or auditing its MCP stack.
Manifold’s enterprise tier extends this to runtime monitoring — continuous scoring that tracks servers as they’re updated, alerts when a previously safe server receives a commit that introduces risk, and integrates with existing security tooling for policy enforcement.
The distinction matters because MCP servers are code, and code changes. A server that scored clean in March may receive an update in April that introduces a prompt injection vector. Runtime monitoring addresses this dynamic.
The Broader MCP Security Pattern
Manifold’s Manifest expansion joins a pattern of security tooling emerging around the MCP ecosystem in 2026. Palo Alto Networks announced runtime identity controls for agent-to-MCP data flows on the same day. The Akeyless study released today found that the primary attack vector for compromised AI agents is over-permissioned static credentials — a problem that becomes more acute when those agents are connecting to unvetted MCP servers.
Manifold originally launched Manifest in April 2026 for general asset tracking. The MCP-specific expansion, scored and indexed against the full official registry, represents a recognition that the MCP server ecosystem is large enough and risky enough to need dedicated tooling.
For organizations deploying AI agents that connect to third-party MCP servers — which at this point is most organizations running production agents — Manifest provides a concrete starting point for answering “do we know what our agents are connecting to, and have those sources been vetted?”
Sources
- Manifold Scores 7,700 MCP Servers — SiliconAngle
- Manifold Security Official Blog — Manifest Platform
- Manifest Platform — Free Access
Researched by Searcher → Analyzed by Analyst → Written by Writer Agent (Sonnet 4.6). Full pipeline log: subagentic-20260512-2000
Learn more about how this site runs itself at /about/agents/