The Model Context Protocol (MCP) ecosystem is formalizing a public registry for server discovery. Earlier this month, the MCP team launched a preview of the official MCP Registry. The Linux Foundation accepted Solo.io’s Agentgateway project into its portfolio. Together they aim to make finding, governing, and running agentic tools more routine for engineering teams.
MCP gives IDEs and agents a standard way to talk to tools and data. As the docs put it, “Think of MCP like a USB-C port for AI applications,” which captures the goal of uniform connectors with minimal client glue. The registry sits above that protocol, publishing a machine-readable catalog of servers that clients can query and install. “We’re launching…the MCP Registry—an open catalog and API for publicly available MCP servers,” the maintainers wrote in the preview announcement.
The preview emphasizes federated discovery rather than a single walled list. The team describes the official service as a “primary source of truth” that public marketplaces and private enterprise sub-registries can mirror and extend, all against a shared OpenAPI. There’s a moderation process (including deny-listing) and, for now, a clear caveat about breaking changes and no durability guarantees during the preview period. The code and API surface are open, with the reference implementation and schema published on GitHub and a hosted API for client integration.
In parallel, the Linux Foundation’s Agentgateway project positions itself as a data plane for agentic systems. It’s an AI-native proxy intended to govern agent-to-agent, agent-to-tool, and agent-to-LLM interactions, with support for emerging protocols such as A2A and MCP. In the Foundation’s words, the project “provides a centralized and secure management layer for AI agent interactions,” bringing policy and observability to an area where general-purpose API gateways are being stretched.
For developers, the near-term flow is straightforward. MCP clients can query the registry API, filter by capability or policy, then install servers without copying manifests by hand. Teams that prefer containerized distribution can also follow Docker’s MCP packaging and registry efforts to reduce workstation variance. Meanwhile, Agentgateway offers a single ingress where platform teams can enforce authN/Z, rate limits, and request inspection across heterogenous agent frameworks while emitting telemetry via OpenTelemetry.
There are notable trade-offs with the registry. Centralized discovery raises questions about curation, governance, and schema drift across public and private sub-registries. During preview, the MCP Registry explicitly makes no durability guarantees and may ship breaking changes, so client code should be defensive. The MCP roadmap calls the registry an API substrate that third-party marketplaces can build on, which implies tighter versioning guarantees, richer metadata, and standardized hooks for provenance and policy.
Agentgateway is also early in its Foundation lifecycle; operators will need to validate performance, protocol coverage, and the maturity of RBAC and auditing in their environments before relying on it for broad policy enforcement. On the gateway side, the Foundation highlights alignment with community specifications (A2A, MCP) and a path toward neutral, vendor-agnostic governance; integrations and production hardening should accelerate as more contributors join.
“Agentgateway is the first and only data plane built from the ground up for AI agents,” the Foundation says; combined with MCP’s “open catalog and API,” the two projects sketch a path to safer scale. For developers looking to learn more explore the MCP Registry repository and agentgateway repository.
