MCP Server(s) Discoverability

[kidehen]

Problem
Discoverability of MCP servers currently relies on a growing number of registries. This approach is not scalable and inevitably leads to a dominant registrar that risks becoming a market choke point.

Solution
Adopt an IANA-registered link relation embedded in the HTML documents of MCP server providers. Similar to how RSS and Atom enabled federated content discovery during the blogosphere era, this method supports a loosely coupled, registry-independent approach to MCP server discovery.

Implementation
The following entry in an MCP server provider’s HTML page is all that’s needed to support this as a recommended best practice:

<link rel="service" href="https://api.example.com/mcp">

[jonathanhefner]

There are ongoing efforts to create an official MCP registry: https://github.com/modelcontextprotocol/registry.

Independent of those efforts, I've thought it could be neat to define a well-known mcp.json URI akin to llms.txt. It could use the same format as the registry's seed.json. Though there would be trust / security issues that need to be addressed.

[kidehen]

Note, the approach I am suggesting can co-exist with others. It's an already tried and test pattern that was used to bootstrap the blogosphere :)