{
  "markdown": "# Entra.Chat Podcast MCP Server\n\n[![npm version](https://img.shields.io/npm/v/entra-news-podcast-mcp)](https://www.npmjs.com/package/entra-news-podcast-mcp)\n[![npm downloads](https://img.shields.io/npm/dt/entra-news-podcast-mcp)](https://www.npmjs.com/package/entra-news-podcast-mcp)\n[![MIT License](https://img.shields.io/badge/license-MIT-blue)](LICENSE)\n\nAn MCP (Model Context Protocol) server for searching transcripts of **[Entra.Chat](https://www.youtube.com/playlist?list=PL06Jj3_onEzEBGRfA7Zddg1IrjgpU1eGp)** — Merill Fernando's Microsoft Entra podcast on YouTube. Ask your AI assistant about Entra ID features, community tools, and identity topics discussed on the show, and get answers with **timestamped YouTube links** that jump straight to the relevant moment.\n\nCompanion to [entra-news-mcp](https://github.com/darrenjrobinson/EntraNewsMCPServer) (the written Entra.News newsletter archive) and [microsoft-ai-roundup-mcp](https://github.com/darrenjrobinson/MicrosoftAIRoundupMCPServer).\n\n## How it works\n\n- **Zero per-user infrastructure.** Install via NPX; the transcript database (SQLite) is downloaded automatically from this repo's GitHub Releases on first run and cached locally (`~/.entra-news-podcast-mcp/`). Updates are checked weekly.\n- **Hybrid retrieval.** BM25 keyword search (in-memory inverted index with Porter stemming — ideal for exact names like \"PIM for Groups\" or \"Maester\" in messy spoken-word transcripts) fused with OpenAI semantic vector search via Reciprocal Rank Fusion. No API key? Keyword search works out of the box.\n- **Timestamped deep links.** Every search result links to `youtube.com/watch?v=...&t=...` so you can hear the discussion in context.\n- **Guest knowledge.** Guests are extracted per episode with their profile links (LinkedIn, Twitter/X, GitHub, Bluesky) — ask \"which episodes was X on?\" or \"who has been on the show?\".\n- **Fresh weekly.** A GitHub Action re-ingests the playlist every week and publishes an updated database release.\n\n## Installation\n\nRequires Node.js **22+** (uses the built-in `node:sqlite` — no native dependencies).\n\n### Claude Desktop\n\nAdd to `claude_desktop_config.json`:\n\n```json\n{\n  \"mcpServers\": {\n    \"entra-podcasts\": {\n      \"command\": \"npx\",\n      \"args\": [\"-y\", \"entra-news-podcast-mcp\"],\n      \"env\": {\n        \"OPENAI_API_KEY\": \"sk-...\"\n      }\n    }\n  }\n}\n```\n\nThe `OPENAI_API_KEY` is **optional** — it enables semantic/hybrid search (embedding the query costs a fraction of a cent). Without it, BM25 keyword search is used.\n\n### Claude Code\n\n```bash\nclaude mcp add entra-podcasts -- npx -y entra-news-podcast-mcp\n```\n\n### VS Code (GitHub Copilot)\n\n```json\n{\n  \"servers\": {\n    \"entra-podcasts\": {\n      \"command\": \"npx\",\n      \"args\": [\"-y\", \"entra-news-podcast-mcp\"]\n    }\n  }\n}\n```\n\n### Cursor\n\n```json\n{\n  \"mcpServers\": {\n    \"entra-podcasts\": {\n      \"command\": \"npx\",\n      \"args\": [\"-y\", \"entra-news-podcast-mcp\"]\n    }\n  }\n}\n```\n\n## Tools\n\n| Tool | Description |\n|------|-------------|\n| `search_entra_podcasts` | Search all episode transcripts. Modes: `hybrid` (default, BM25 + semantic via RRF), `semantic`, `keyword`. Results include episode, guests, and timestamped YouTube links. |\n| `get_episode` | Full episode by `video_id` or `date` — metadata, guests with profile links, chapters with timestamped links, and the complete transcript with `[mm:ss]` markers. |\n| `list_episodes` | Browse the archive; filter by `year`, `month`, and/or `guest` name. |\n| `list_guests` | Directory of all podcast guests with profile links, appearance counts, and latest appearance. |\n| `get_guest` | One guest by name: profile links and **every episode they appeared on**. |\n| `find_tool_mentions` | Community tools discussed on the show, with timestamped links to hear the discussion. |\n\n### Example prompts\n\n- *\"What did Merill and his guests say about PIM for Groups?\"*\n- *\"Which episodes has Jane Doe been on? What's her LinkedIn?\"*\n- *\"Find where Maester was discussed and give me the YouTube timestamp.\"*\n- *\"List the Entra.Chat episodes from March 2026.\"*\n\n## Data & freshness\n\n| | |\n|---|---|\n| Source | [Entra.Chat playlist](https://www.youtube.com/playlist?list=PL06Jj3_onEzEBGRfA7Zddg1IrjgpU1eGp) on YouTube (@merillx) |\n| Transcripts | YouTube captions (uploaded captions preferred, auto-generated otherwise) |\n| Refresh | Weekly GitHub Action → new `db-*` release |\n| Local cache | `~/.entra-news-podcast-mcp/` (`%USERPROFILE%\\.entra-news-podcast-mcp\\` on Windows) — update check at most every 7 days; delete the folder to force a fresh download |\n| Runtime override | Set `ENTRA_PODCAST_DB_PATH` to use a local database file instead |\n\n## Development\n\n```bash\nnpm install\nnpm run build\n\n# Ingest (requires yt-dlp on PATH: pipx install yt-dlp / winget install yt-dlp)\nnode dist/scripts/ingest.js --limit 2        # test with 2 videos\nnode dist/scripts/ingest.js                  # full playlist backfill\nnode dist/scripts/ingest.js --incremental    # only new videos\nnode dist/scripts/ingest.js --video <id>     # one video (re-ingests)\nnode dist/scripts/ingest.js --reextract      # rebuild guests + tool mentions from stored\n                                             # data (no network) — after editing\n                                             # guest-overrides.json or known-tools.ts\nnode dist/scripts/ingest.js --embed-missing  # embed chunks that have no embedding yet\n                                             # (needs OPENAI_API_KEY; no YouTube access)\n\n# Run the server against a local DB\nENTRA_PODCAST_DB_PATH=./entra-news-podcasts.db npx @modelcontextprotocol/inspector node dist/src/index.js\n```\n\nSet `OPENAI_API_KEY` during ingest to generate embeddings (semantic search); without it the ingest still completes and BM25 keyword search works.\n\n### Guest extraction overrides\n\nGuests are extracted heuristically from video titles/descriptions. Episodes the heuristics miss are listed at the end of each ingest run — correct them in [scripts/lib/guest-overrides.json](scripts/lib/guest-overrides.json) (keyed by `video_id`, entries fully replace extraction for that video) and apply with `node dist/scripts/ingest.js --reextract` (no re-download needed).\n\n### Manual database refresh (if YouTube blocks CI)\n\nYouTube sometimes blocks caption downloads from datacenter IPs (`BOT_BLOCKED` in the workflow log). Consumers are unaffected — the last-good release stays `latest`.\n\nTwo mitigations are built in before falling back to a manual refresh:\n\n- CI installs [Deno](https://deno.com/), which yt-dlp requires as a JS runtime to solve YouTube's player challenges — without it, requests are far more likely to be flagged as bot traffic.\n- The `YTDLP_PROXY` repository secret (set since 2026-07-21 to a residential proxy URL in `http://user:pass@host:port` form — note IPRoyal's dashboard shows `host:port:user:pass`, which must be rewritten) routes all yt-dlp traffic through that proxy. The same env var works for local ingest runs. If the proxy account runs out of traffic, top it up or clear the secret and refresh manually.\n\nIf CI is still blocked, refresh manually from a residential IP:\n\n```bash\nnode dist/scripts/ingest.js --incremental\ngh release create \"db-$(date -u +%Y.%m.%d)-9999\" \\\n  --repo darrenjrobinson/EntraNewsPodcastMCPServer \\\n  --title \"Database Update (manual)\" --latest \\\n  ./entra-news-podcasts.db\n```\n\n## Release process (maintainer)\n\nCode releases (`v*` tags) publish to npm (Trusted Publishing / OIDC, no tokens) and the MCP Registry (`io.github.darrenjrobinson/entra-news-podcast`) via `.github/workflows/publish-mcp.yml`:\n\n1. Bump `version` in `package.json` **and both version fields in `server.json`** (CI enforces lockstep).\n2. Commit, then `git tag v0.x.y && git push origin main v0.x.y`.\n\nDatabase releases (`db-*` tags) are produced by the weekly workflow and never trigger an npm publish.\n\n## License\n\nMIT © [Darren Robinson](https://github.com/darrenjrobinson)\n",
  "bytes": 7916,
  "sha": "ee1fc150e9ecda69178353c42d690aacacb368ad11e69eef02192ffa18dd332e",
  "repo_slug": "darrenjrobinson/entranewspodcastmcpserver",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/mcp_io_github_darrenjrobinson_entra_news_pod_37087520/readme"
}