Skip to content

feat: Explore WebMCP support for agent-accessible text references #52

Description

@maehr

Foundation landed (2026-08-25)

The two read-only capabilities this issue lists first now exist as plain HTTP, so a WebMCP surface can wrap them instead of inventing them:

Validating that a reference exists and finding its resolver targets both follow from a /id/ref/{uuid}.json fetch. Scope the exploration to the tool definitions and the transport, and treat the data surface as settled.


Summary

Explore how TextRefs could use WebMCP to make its registry and canonical references directly accessible to browser-based AI agents.

Motivation

TextRefs provides stable identifiers, canonical citation data, and links to editions of referenced works. Browser agents currently need to interpret pages and links themselves, which can lead to unreliable reference resolution.

WebMCP could allow TextRefs to expose its core capabilities more explicitly, helping agents:

  • Search for registered works and citation systems
  • Resolve a traditional citation to a canonical TextRefs reference
  • Validate whether a reference exists
  • Find available editions or resolver targets
  • Retrieve structured information about the current reference

Proposed change

Define a small, read-only WebMCP surface that reflects TextRefs’ existing public data and use cases.

The initial scope should prioritize:

  1. Searching the registry
  2. Resolving canonical references
  3. Listing available resolver targets

WebMCP support should complement the existing HTML and JSON-LD resources rather than replace them.

Considerations

  • Treat WebMCP as a progressive enhancement
  • Preserve stable URLs and structured-data access for non-browser integrations
  • Keep the initial scope read-only
  • Ensure results clearly distinguish valid citation syntax from references that are actually registered
  • Avoid requiring WebMCP for normal use of the site
  • Checkout stateless MCPs https://modelcontextprotocol.io/seps/2575-stateless-mcp

Alternatives considered

No response

Acceptance criteria

  • The intended agent workflows are documented
  • The initial read-only capabilities are agreed
  • The relationship between WebMCP, HTML pages, and JSON-LD resources is clear
  • Experimental browser support and compatibility risks are acknowledged
  • Follow-up work can be split into smaller implementation issues

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or requestpost-v0.1.0Deferred past the v0.1.0 baseline. Revisit if the need arises.questionFurther information is requested

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions