io.github.darrenjrobinson/entra-news

entra-news-mcp

A searchable knowledge MCP over Entra.news — Merill Fernando's curated weekly digest of Microsoft Entra news, features, and community tools.

What is this?

Entra.news is a high-signal, curated newsletter covering Microsoft Entra (Azure AD) features, announcements, and community tools — published weekly since mid-2023.

This MCP server exposes the full historical archive as a natural language search interface. Ask questions and get sourced answers directly from past issues — including issue number, date, and canonical URL.

Zero per-user infrastructure. Users install an NPX package. That's it.

Quick Start

Claude Desktop

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS, %APPDATA%\Claude\claude_desktop_config.json on Windows):

{
  "mcpServers": {
    "entra-news-mcp": {
      "command": "npx",
      "args": ["entra-news-mcp"]
    }
  }
}

Restart Claude Desktop. The database (~15–20 MB) will be downloaded on first launch and cached in ~/.entra-news-mcp/.

Cursor / Copilot Studio / Any MCP Host

{
  "mcpServers": {
    "entra-news-mcp": {
      "command": "npx",
      "args": ["-y", "entra-news-mcp"]
    }
  }
}

VS Code

Add to your MCP config (workspace .vscode/mcp.json, or run MCP: Add Server from the Command Palette):

{
  "servers": {
    "entra-news-mcp": {
      "command": "npx",
      "args": ["-y", "entra-news-mcp"]
    }
  }
}

Semantic Search (Optional)

By default the server uses keyword and phrase matching over the SQLite index — no API key needed. For significantly better result quality, set your OpenAI API key to enable semantic + hybrid search:

{
  "mcpServers": {
    "entra-news-mcp": {
      "command": "npx",
      "args": ["entra-news-mcp"],
      "env": {
        "OPENAI_API_KEY": "sk-..."
      }
    }
  }
}

Local Database Cache

On first launch the server downloads the database (~15–20 MB) from GitHub Releases and caches it locally:

The server checks for a newer database release once per week. If you want to force an immediate re-download (e.g. after a new issue has been ingested), delete the cache folder and restart your MCP host:

Windows (PowerShell):

Remove-Item "$env:USERPROFILE\.entra-news-mcp" -Recurse -Force

macOS / Linux:

rm -rf ~/.entra-news-mcp

Available MCP Tools

Example queries

• "What did Entra.news cover about Conditional Access in 2024?"
• "Show me the issue from March 2025"
• "What PowerShell tools for Entra have been mentioned?"
• "Has there been coverage of Verified ID?"
• "List all issues from 2024"

Architecture

Substack API (entra.news/api/v1/posts)
    │
    ▼
Node.js ingestion script  ←  OpenAI text-embedding-3-small
    │
    ▼
SQLite (chunks + embeddings, ~15–20 MB)
    │
    ▼
GitHub Release asset  ──→  NPX MCP Server
                               └─ Downloads DB on first run
                               └─ Checks for updates weekly
                               └─ In-memory vector similarity + keyword search

Cost: ~$0.01/week (embeddings on new issues only). Zero hosting.

Running the Ingestion Pipeline

Note: You only need to do this if you're maintaining your own fork or building the initial index. End users just run npx entra-news-mcp — the database is downloaded automatically.

Prerequisites

• Node.js 22+
• An OpenAI API key (text-embedding-3-small access)

Full ingest (first time)

# Set your API key
$env:OPENAI_API_KEY = "sk-..."

# Run the ingestion pipeline
./scripts/ingest.ps1

Or directly with Node.js:

export OPENAI_API_KEY=sk-...
npm install && npm run build
node dist/scripts/ingest.js

Incremental update (new issues only)

./scripts/ingest.ps1 -Incremental

node dist/scripts/ingest.js --incremental

The output database (entra-news.db) should then be uploaded as a GitHub Release asset — the GitHub Actions workflow handles this automatically on a weekly schedule.

Automated Weekly Updates

A GitHub Actions workflow (.github/workflows/weekly-update.yml) runs every Monday at 9am Sydney time (Sunday 23:00 UTC), shortly after each new Entra.news issue is published:

1. Downloads the current database from GitHub Releases
2. Runs the incremental ingestion pipeline
3. Publishes the updated database as a new GitHub Release

Required secret: Add OPENAI_API_KEY to your repository secrets (Settings → Secrets).

Releasing (npm + MCP Registry)

Publishing is automated by .github/workflows/publish-mcp.yml, triggered by pushing a v* tag. Authentication is tokenless (OIDC) for both npm (Trusted Publishing) and the MCP Registry.

1. Bump the version in package.json and server.json (both version fields) — the workflow fails if they don't match the tag
2. Commit, then tag and push — deriving the tag from package.json so it always matches the workflow's version gate:

VERSION=$(node -p "require('./package.json').version")
git tag "v$VERSION"
git push origin main "v$VERSION"

The workflow builds, publishes to npm (with provenance), and publishes the new version to the MCP Registry.

Development

npm install
npm run build          # Compile TypeScript
npm start              # Run the MCP server

Project structure

src/
  index.ts             # Entry point
  server.ts            # MCP server + tool registration
  db/
    client.ts          # SQLite client — DB download/cache + search
  tools/
    search.ts          # search_entra_news tool
    get-issue.ts       # get_issue tool
    list-issues.ts     # list_issues tool
    find-tool-mentions.ts  # find_tool_mentions tool
  utils/
    embeddings.ts      # OpenAI embedding helper
scripts/
  ingest.ts            # Full ingestion pipeline (TypeScript)
  ingest.ps1           # PowerShell wrapper for ingestion
.github/workflows/
  weekly-update.yml    # Automated weekly update

Permissions & Content

The Entra.news content is © Merill Fernando & Joshua Fernando. This tool accesses the publicly available Substack API (not scraping) and is intended for personal/community use. Please reach out to hey@entra.news before any public deployment.

Author

Built by Darren Robinson.

Entra.news by Merill Fernando.