mcpField
D
discourse

discourse

github.com/atholaProductivity34381 installsUpdated 3d ago
Repository

About discourse

A research skill for gathering and comparing community discussions about technologies or approaches. It searches at least two channels, including Hacker News, Lobsters, Reddit, or selected tech blogs, and retains source attribution.

Key features

  1. 01Search Hacker News via Algolia
  2. 02Search Lobsters, Reddit, and tech blogs
  3. 03Combine findings with source attribution
  4. 04Report failures instead of fabricating results

Use cases

  1. 01Gather community opinions on a technology
  2. 02Find experience reports from HN, Reddit, or Lobsters

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/athola/claude-night-market --skill discourse

After installing, type/discoursein Claude Code to confirm it loaded.

Details

From the project README

Discourse Search

When To Use

  • Gathering community opinions on a technology or approach
  • Finding experience reports from HN, Reddit, or Lobsters

When NOT To Use

  • Academic research (use Skill(tome:papers))
  • Code examples (use Skill(tome:code-search))

Scan community channels for discussions on a topic.

Channels

  • Hacker News: Algolia API at hn.algolia.com
  • Lobsters: WebSearch with site:lobste.rs
  • Reddit: JSON API (append .json to URLs)
  • Tech blogs: WebSearch targeting curated domains

Workflow

  1. Build search URLs/queries per channel using tome.channels.discourse.* functions
  2. Execute via WebFetch (APIs) or WebSearch (fallback)
  3. Parse responses into Finding objects
  4. Merge across sources with source attribution

Exit Criteria

  • At least two community channels (HN, Lobsters, Reddit, or tech blogs) queried per invocation
  • HN results fetched via Algolia API at hn.algolia.com (WebFetch); WebSearch used as fallback if the API is unreachable
  • Each Finding object includes a source field identifying which channel (HN, Lobsters, Reddit, or blog) it came from
  • If all WebFetch and WebSearch calls fail for every channel, the failure is reported explicitly rather than returning fabricated or empty findings

Comments

Markdown supported