Manifold scores 7,700 MCP servers in Manifest expansion aimed at agent security teams

Artificial intelligence detection and response platform startup Manifold Security Inc. today announced an expansion of its Manifest supply chain intelligence tool to cover Model Context Protocol servers, adding scored entries for more than 7,700 servers pulled from the official MCP Registry.

The expansion brings MCP servers into the same searchable index Manifold already maintains for agent skills and plugins. Each server gets a composite Manifest Score built from two signal families: a Lineage Score that grades publisher provenance and a Safety Score that flags behavioral risk in the server’s declared surface. Higher scores indicate greater confidence in the component’s provenance and hygiene.

MCP, the protocol developed by Anthropic PBC, has become the de facto plumbing connecting AI agents to external tools and data sources. Adoption has been widespread, with support now built into most major agent platforms, but the security review process around individual servers has not kept pace – the security gap is the problem the new release from Manifold is targeting.

Unlike agent skills, which are typically markdown files backed by a public GitHub repository, many MCP servers expose only an HTTP endpoint. The tool descriptions and metadata declared at that endpoint are often the only surface a security team can inspect. A server can accept inputs from an agent and forward them to a third party with no indication in the declared interface that it does so, leaving security teams reliant on trust rather than verification.

A compromised MCP server does not just shape an agent’s reasoning, it controls execution and data flow, making provenance and behavioral signals material to any enterprise rolling out agentic workflows.

The Lineage Score draws on authorship history, community footprint, repository age, commit patterns and any naming or identity discrepancies between a registry listing and its underlying source. The Safety Score inspects the server’s declared interface for content that contradicts its stated behavior or attempts to manipulate the calling agent, including coercive instructions aimed at overriding agent decision-making and tool descriptions that contain embedded prompt injection.

Manifest is offered in two tiers. The open-access free tier provides the full index of scored servers with Lineage, Safety and composite Manifest Scores, searchable alongside the existing skills and plugins catalog. An enterprise tier integrates directly with the Manifold platform, automatically enriching every agent-connected MCP server discovered in a customer environment with its Manifest Score.

The company has now indexed more than 206,000 assets across skills, plugins and MCP servers from 31,472 unique publishers, with additional registry sources and GitHub-hosted servers under evaluation for future coverage.

Manifold is a venture capital-backed startup that was in the news in March when it raised $8 million in funding to fuel development of its platform. Investors in the company include Costanoa Ventures, Cherry Ventures Management GmbH, Rain Capital, Modern Technical Fund and a number of individual investors.

Image: Manifold Security