Skip to main content
Glama
JW MCP Server
by advenimus

JW MCP Server

npm version GitHub Release

An MCP server for working with JW.org content — Bible scripture lookup with study notes, workbook materials, Watchtower articles, and video captions.

Quick Start

claude mcp add jw-mcp -- npx -y jw-mcp

Claude Desktop

Download the latest jw-mcp.mcpb from Releases and open it — Claude Desktop will install it automatically.

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

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

Other MCP Clients (Cursor, Windsurf, etc.)

npx -y jw-mcp

Related MCP server: jw-org-mcp

Docker (self-hosted)

Run this server on a machine with a public HTTPS URL, then paste that URL into Grok, Claude.ai, or ChatGPT as a custom connector.

Compose binds the container to loopback only (127.0.0.1:8080). Do not publish port 8080 on the public internet. Put a reverse proxy on the host, terminate TLS there, and proxy to 127.0.0.1:8080.

Use port 443. Claude.ai never connects to a custom connector on any other port. It only says "Couldn't reach the MCP server". Grok accepts other ports, so a URL like https://host:18443/mcp can work in Grok and still fail in Claude. The server logs a warning at startup when MCP_BASE_URL has a port.

  1. Copy .env.example to .env.

  2. Set MCP_BASE_URL to your public origin, with no path and no port. Example: https://jw-mcp.example.com.

  3. Set MCP_AUTH_SECRET to a long access key (at least 16 characters).

  4. Keep MCP_AUTH=true. HTTP mode with auth off is loopback-only. Do not disable auth in Docker.

  5. Set MCP_TRUST_PROXY=true only when exactly one proxy sits in front of the server and sets or appends the real client IP in X-Forwarded-For (Caddy, nginx, or a Cloudflare Tunnel). Leave it unset otherwise.

  6. Start it:

docker compose up --build -d

The connector URL is:

https://your-host/mcp

When you add the connector, the AI site opens a login page. Type the same access key you put in MCP_AUTH_SECRET.

Client

Where to add it

Grok

grok.com/connectors → New Connector → Custom

Claude.ai

Customize → Connectors → Add custom connector

ChatGPT

Settings → enable Developer mode → add the server URL

Local HTTP (http://localhost:8080) is fine for testing because compose binds loopback. Claude, ChatGPT, and Grok must reach a public HTTPS URL. A tunnel such as ngrok or Cloudflare Tunnel can expose that loopback port for a live click-test. Do not publish 8080 on 0.0.0.0.

Auth codes, clients, and tokens are stored in the jw-mcp-auth Docker volume so a container restart does not drop every connection.

Home hosting with Cloudflare Tunnel

Use this when the server runs on a home or office network. The tunnel makes an outbound-only connection to Cloudflare, so you open no router ports and your home IP stays out of DNS. Requests to jw.org still leave from your own network. Your domain's DNS must be on Cloudflare.

  1. In Cloudflare, go to Zero Trust → Networks → Tunnels and create a tunnel (type Cloudflared). Copy its token.

  2. In the tunnel's Public hostname tab, add your hostname (for example jw-mcp.example.com) with service http://jw-mcp:8080. If a DNS record for that name already exists, delete it first.

  3. In .env, set:

    MCP_BASE_URL=https://jw-mcp.example.com
    MCP_TRUST_PROXY=true
    CLOUDFLARE_TUNNEL_TOKEN=<token from step 1>
  4. Start both containers:

    docker compose --profile tunnel up --build -d
  5. Remove any old router port forward or reverse-proxy site for this server. Nothing needs to reach it from the internet directly.

If the tunnel doesn't come up, run docker compose logs cloudflared. A missing or wrong CLOUDFLARE_TUNNEL_TOKEN makes it restart over and over.

Cloudflare settings that can block AI connectors. Check them for this hostname:

  • Security → Bots: keep Bot Fight Mode off. On the free plan it can't be skipped for one hostname, and it can challenge server-to-server calls from Claude and Grok.

  • AI Crawl Control / Block AI bots: don't block Claude or Grok user agents on this hostname.

  • Cloudflare Access: don't put an Access login in front of this hostname. Claude's servers can't pass it. The server's own access key already protects it.

  • If you use WAF rules, allow Anthropic's outbound range 160.79.104.0/21.


Tools

All tools support multiple languages via the langwritten parameter (E = English, S = Spanish, F = French, etc.).

Bible Scripture Tools

search_bible_books

Search for Bible books by name, abbreviation, or number.

{ "query": "matthew" }

get_bible_verse

Get plain verse text from wol.jw.org. Books are numbered 1-66 (1-39 OT, 40-66 NT).

{ "book": 43, "chapter": 3, "verse": 16 }

get_verse_with_study

Get verses with study notes, cross-references, and research articles. Supports ranges.

{
  "book": 40, "chapter": 5, "verse": "3-5",
  "fields": ["verses", "study_notes", "study_articles"]
}

Available fields: verses, study_notes, study_articles, cross_references, chapter_level, combined_text

Scripture Tools Demo

get_bible_verse_url

Generate JW.org URLs for verses, ranges, or chapters — useful for adding clickable links to documents.

{ "book": 19, "chapter": 83, "verse": "18" }
{ "book": 23, "chapter": 46, "verse": "9-11" }
{ "book": 40, "chapter": 5 }

Get Verse URL Example


Workbook Tools

Get available Christian Life and Ministry workbook weeks for the current or a specific issue.

{ "issue": "20250500", "langwritten": "E" }

getWorkbookContent

Download and parse a workbook week's RTF content to clean plain text (70% token reduction).

{ "url": "https://cfp2.jw-cdn.org/a/clm_E_202505_01.rtf" }

Workbook Content Example


Watchtower Tools

Get available Watchtower study articles. Automatically uses the correct issue (published 2 months ahead of study period).

{ "issue": "20250300", "langwritten": "E" }

getWatchtowerContent

Download and parse a Watchtower article's RTF content to clean plain text (70% token reduction).

{ "url": "https://cfp2.jw-cdn.org/a/w_E_202509_01.rtf" }

Watchtower Content Example


Video Caption Tools

get_jw_captions

Fetch video captions and metadata by video ID or any JW.org URL.

{ "video_id": "pub-jwbvod25_17_VIDEO" }

Also accepts full JW.org URLs — the video ID is extracted automatically.

Video Captions Example


Troubleshooting

  1. "Could not read package.json" — Use absolute path in Claude Desktop config

  2. Server disconnects — Ensure Node.js is installed (npm install)

  3. Permission denied — chmod +x start-server.sh

Development

npm start          # stdio mode (local)
npm run start:http # HTTP mode (testing)
npm test           # OAuth + HTTP connector tests

Built with Node.js, MCP SDK, node-fetch, and cheerio.

License

MIT

Available Tools

9 tools
get_bible_verseA

Get plain Bible verse text from wol.jw.org. Returns just the verse text without study notes or additional content. For comprehensive study content including notes and cross-references, use get_verse_with_study instead.

Inputs: book (required), verse (required), chapter (required).

get_bible_verse_urlA

Get the jw.org URL for a Bible verse or range of verses. Returns a direct link to view the scripture on jw.org. Supports single verses (e.g., verse: "18"), verse ranges (e.g., verse: "14-16"), and comma-separated verses (e.g., verse: "1,3,5" - will convert to range if contiguous). Use search_bible_books to find book numbers. Perfect for adding clickable scripture links to markdown documents.

Inputs: book (required), chapter (required), verse.

get_jw_captionsA

Fetches video captions from JW.org by video ID or URL. Accepts either a direct video ID (e.g., "pub-jwbvod25_17_VIDEO") or a JW.org URL (e.g., "https://www.jw.org/finder?srcid=jwlshare&wtlocale=E&lank=pub-jwbvod25_17_VIDEO")

Inputs: video_id (required).

get_verse_with_studyB

Get Bible verse(s) with comprehensive study content from wol.jw.org. Supports single verses or ranges (e.g., "14-16"). Returns verse text, study notes, cross-references, research articles from the Research Guide, and chapter outlines. Field selection allows you to customize what content is returned.

Inputs: book (required), verse (required), chapter (required), fetch, +2 more.

getWatchtowerContentA

STEP 2: Get the actual Watchtower article content after user chooses an article. Use this tool AFTER getWatchtowerLinks when user specifies which article they want (e.g., "Imitate the Faithful Angels" or "Look to Jehovah for Comfort"). Takes the RTF URL from Step 1 results, downloads the RTF file, parses it to clean plain text, and returns the formatted article content with proper structure and line breaks.

Inputs: url (required).

getWorkbookContentA

STEP 2: Get the actual CLM workbook content after user chooses a week. Use this tool AFTER getWorkbookLinks when user specifies which week they want (e.g., "May 5-11" or "June 30-July 6"). Takes the RTF URL from Step 1 results, downloads the RTF file, parses it to clean plain text, and returns the formatted workbook content with proper line breaks and structure.

Inputs: url (required).

search_bible_booksA

Search for Bible books by name or abbreviation. Returns book numbers (1-66) and names. Useful for finding the correct book number for other scripture tools. Examples: "matt" -> Matthew (40), "1 john" -> 1 John (62), "gen" -> Genesis (1).

Inputs: query (required), limit.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.8/5.0

    Scored across 9 tools

    Disambiguation4/5

    Most tools have distinct purposes, but there is some overlap between get_bible_verse and get_verse_with_study, as both retrieve Bible verses, though the latter includes study content. The descriptions clearly differentiate them, reducing confusion. Other tools target unique resources like captions, Watchtower articles, or workbook content, making them easy to distinguish.

    Naming Consistency2/5

    Naming is inconsistent, mixing snake_case (e.g., get_bible_verse) with camelCase (e.g., getWatchtowerContent) and varying verb styles. Some tools use 'get' prefix, while others use 'search', but the pattern is not uniform across all tools. This inconsistency can make the tool set harder to navigate and predict.

    Tool Count5/5

    With 9 tools, the count is well-scoped for a JW.org content server, covering Bible verses, study materials, Watchtower articles, and meeting workbooks. Each tool serves a specific function, and there are no extraneous tools, making the set manageable and focused on the domain.

    Completeness4/5

    The tool set covers key areas like Bible access, study content, and publication retrieval, with clear workflows (e.g., getWatchtowerLinks followed by getWatchtowerContent). Minor gaps might include direct search for publications or additional Bible study features, but core operations are well-covered, allowing agents to perform most tasks without dead ends.

    Maintenance

    ActivityMaintained
    ResponsivenessLimited data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      Provides read-only access to the Watchtower Online Library (WOL) with advanced search capabilities, document retrieval, and multilingual support. Enables users to search and browse Jehovah's Witnesses publications through natural language interactions.
      5
      MIT
    • A
      license
      A
      quality
      C
      maintenance
      An MCP server that provides controlled, verifiable access to official jw.org content, enabling AI applications to search articles, retrieve full articles, and lookup scriptures without hallucinations.
      4
      16
      GPL 3.0
    • A
      license
      A
      quality
      D
      maintenance
      A remote MCP server for traceable research across JW.org and the Watchtower Online Library, providing tools for searching, retrieving articles, footnotes, cross-references, and daily texts with optional AI-powered synthesis.
      20
      2
      MIT