Skip to content

contribution-guide.md

Latest commit

 

History

History
62 lines (44 loc) · 3.61 KB

contribution-guide.md

File metadata and controls

62 lines (44 loc) · 3.61 KB
Raw

MCP Index Metadata Contribution Guide

TensorBlock MCP Index turns this awesome list into structured data that agents and users can search, compare, and install from.

The indexer reads the markdown catalog and generates:

  • structured catalog entries,
  • server profile JSON,
  • install config previews,
  • and a local MCP registry that agents can query directly.

When adding a server, include as much of this metadata as possible in the issue form, PR body, or entry description:

  • Install command: npx, uvx, pip, Docker, or remote endpoint.
  • Transport: stdio, streamable-http, or sse.
  • Auth: none, API key, OAuth, bearer token, or other.
  • Supported clients: Claude Desktop, Cursor, Codex, VS Code, or other MCP clients.
  • Tool count and important tool names.
  • License.
  • Docs URL.
  • Remote MCP endpoint, if public.

For short markdown entries, structured metadata can live in data/server-metadata/{serverId}.json. Server submissions created through the Add MCP server issue form can generate this sidecar automatically, so the category page stays readable while the profile/API still get install, transport, auth, client, and license fields.

Server submission issues are automatically tagged with intake status labels:

  • needs-metadata: required fields, URL validation, or category routing still need attention.
  • duplicate: the submitted project URL already matches an existing catalog entry.
  • ready-for-pr: automation created or updated a draft docs PR for maintainer review.
  • automation-blocked: automation generated a branch, but GitHub permissions blocked PR creation.

The manual MCP Add Server Intake Refresh workflow can backfill these labels for old open submissions. It defaults to dry-run mode so maintainers can inspect the planned label changes before applying them.

For existing indexed servers, use the metadata improvement issue form with the TensorBlock profile id and structured values such as:

Install: npx -y example-mcp
Transport: stdio
Auth: api-key, requires EXAMPLE_API_KEY
Docs URL: https://docs.example.com/mcp
License: MIT
Tools: search, fetch_profile
Tool count: 2

When the profile id matches the index and the values are clear, automation drafts a metadata sidecar PR for maintainer review.

For duplicate, stale, broken, unsafe, or poorly categorized entries, use the broken entry issue form with the TensorBlock profile URL, server id, or project URL. Clear dead-link reports can generate direct cleanup PRs against the matching docs/*.md category page. Duplicate, stale, category, safety, or unclear reports generate draft investigation PRs under docs/broken-entry-reports/ so maintainers can verify the problem before editing catalog docs or metadata sidecars.

TensorBlock also runs a scheduled catalog health check. It scans the generated catalog for duplicate primary links and rotates through indexed GitHub repositories to detect 404, archived, or disabled projects. New findings open catalog-health issues that feed into the same broken-entry report workflow.

For scoped community work, start from the MCP Index Community Cleanup Queue. It links to live issue queues for good first metadata fixes, broken entries, catalog-health findings, server submissions, client config requests, and profile claims.

Complete metadata helps TensorBlock generate:

  • server profiles,
  • install configs,
  • registry entries,
  • install confidence notes,
  • and future verification reports.

This makes each listed server more useful than a directory link: agents can discover it by intent, inspect profile metadata, and draft a client config without manually reading every README.