Skip to content

Map out the end-to-end discovery path (well-known + skills over MCP + configuration) #79

@olaservo

Description

@olaservo

Context

From the April 7, 2026 Office Hours, there's a need to fully map out the end-to-end discovery path — how does skill discovery actually work from domain resolution through configuration to runtime loading?

Currently, several pieces exist independently:

  • Well-known URIs — the Agent Skills Discovery RFC defines /.well-known/agent-skills/index.json for domain-level discovery. Vercel/Mintlify already supports npx skills add <domain> pulling from well-known URIs.
  • Skills over MCP — the skill:// URI scheme and skill://index.json resource for MCP-level discovery (SEP in progress via Skills Extension SEP — spec, URI scheme, and reference implementation #75).
  • Configuration — how clients are configured to connect to skill servers and load skills at runtime.

What's missing is a clear picture of how these layers compose into a single coherent path.

Deliverable

A document or diagram mapping the complete discovery pathway:

  1. Domain resolution — user provides a domain → well-known URI lookup
  2. Manifest/index retrievalindex.json content and structure
  3. MCP server connection — how the well-known URI leads to an MCP server endpoint
  4. Skill resource loadingskill:// resource discovery and reading at runtime
  5. Client configuration — how this gets wired into a client's settings

Using modelcontextprotocol.io as a dog-fooding example was discussed. Currently Mintlify controls the well-known URI for that domain, so hosting custom MCP skills there would require either Mintlify adding that feature or using a separate subdomain.

Related

🦉 Generated with Claude Code

Metadata

Metadata

Assignees

Labels

action-itemAction item from a meeting or office hoursresearchInvestigation or analysis needed

Type

No type

Projects

Status

In progress

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions