existdb-mcp

An MCP (Model Context Protocol) server that connects Claude to eXist-db — the native XML database widely used in the digital humanities as the backbone of TEI Publisher, scholarly editions, linguistic corpora and institutional archives.

If you work with TEI/XML corpora, manage a DH project, or develop XQuery applications, this server lets Claude:

• read your offline copy of the eXist-db function reference,

• run XQueries against your live database,

• inspect, retrieve, store and delete documents and collections —

all through natural-language prompts in Claude Code or Claude Desktop. No more context-switching between the chat, eXide and the REST endpoint.

What you get

• 📚 Offline documentation lookup — full signatures for every built-in function, served from a local cache of the eXist-db fundocs (no internet needed once scraped)

• ⚙️ Live XQuery execution — run queries against your eXist-db instance via the REST API, with paginated results

• 🗄️ Collection management — list, create, store, retrieve and delete collections and documents (think /db/corpus, /db/apps/edition, …)

• 🧠 XQuery snippets — ready-to-paste templates for common patterns (RESTXQ endpoints, Lucene full-text, XSLT pipelines, HTTP client, …)

Who is this for?

• 👩‍💻 XQuery / eXist-db developers building apps, RESTXQ APIs or data pipelines.

  "Look up xmldb:store, then store this XML as /db/myapp/data/01.xml."

• 📜 Researchers & digital humanists working on TEI editions, corpora or critical editions.

  "Extract all <persName> values from /db/corpus and return a frequency-ordered list."

• 🗂️ Archivists & curators managing XML-encoded collections (EAD, TEI, METS, EpiDoc, …).

  "List every document in /db/archive/finding-aids modified in the last 30 days, with size and last editor."

Table of contents

• Quick start

• Configuring Claude

• Environment variables

• Available tools

• Example prompts

  • Working with TEI corpora

• Ecosystem

• Testing

• Troubleshooting

• Regenerating the documentation cache

• Project structure

• Contributing

• License

Quick start

1. Clone and install

git clone https://github.com/your-username/existdb-mcp.git
cd existdb-mcp
npm install

2. Build the documentation cache

This scrapes exist-db.org/exist/apps/fundocs and writes src/cache/fundocs.json:

npm run scrape

No internet, or want to try the server right away? Use the offline fixture instead — it ships a tiny but real cache so the documentation tools work out of the box:

npm run scrape -- --offline

3. Build the TypeScript

npm run build

The build step compiles src/ to dist/ and copies the documentation cache into dist/cache/ so the server can find it at runtime.

4. Try it

npm start
# > existdb-mcp running on stdio

The server speaks the MCP stdio protocol — point an MCP client at it (see below) to start using the tools.

Configuring Claude

Claude Code

Add to your project's .claude/mcp.json (or ~/.claude/mcp.json for a global install):

{
  "mcpServers": {
    "existdb": {
      "command": "node",
      "args": ["/absolute/path/to/existdb-mcp/dist/index.js"],
      "env": {
        "EXISTDB_ENDPOINT": "http://localhost:8080/exist",
        "EXISTDB_USER": "admin",
        "EXISTDB_PASSWORD": "your-password"
      }
    }
  }
}

Claude Desktop

Add to claude_desktop_config.json:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

• Windows: %APPDATA%\Claude\claude_desktop_config.json

• Linux: ~/.config/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "existdb": {
      "command": "node",
      "args": ["/absolute/path/to/existdb-mcp/dist/index.js"],
      "env": {
        "EXISTDB_ENDPOINT": "http://localhost:8080/exist",
        "EXISTDB_USER": "admin",
        "EXISTDB_PASSWORD": "your-password"
      }
    }
  }
}

Restart Claude Desktop / Claude Code and the existdb tools will appear in the available-tools list.

Tip — docs only, no eXist-db running? You can omit the env block entirely. The documentation tools work offline; only the query/collection tools require a live server.

Environment variables

All three can also be passed inline as arguments to any tool call (endpoint, username, password) — useful when you want Claude to connect to a different instance for one-off queries.

Available tools

Documentation (offline — no eXist-db needed)

Snippet topics: store-document, fulltext-search, restxq, http-client, security, transform-xslt, collection-query.

Query execution (requires a live eXist-db)

Collection management (requires a live eXist-db)

Example prompts

Once connected to Claude, you can ask things like:

Discover the API

What parameters does xmldb:store accept?
List all functions in the Lucene full-text namespace.
Search for functions related to scheduling.
Show me a RESTXQ endpoint snippet.
What namespaces are available for security?

Query a live database

On my local eXist-db, run:
  for $doc in collection("/db/myapp")//record
  return $doc/title/string(.)

Validate this XQuery without running it:
  let $x := 1
  return $x +

Count how many documents are in /db/apps/dashboard.

Manage collections and documents

List the contents of /db/myapp/data.

Create a new collection called "drafts" inside /db/myapp.

Store this XML as /db/myapp/config/settings.xml:
  <settings><theme>dark</theme></settings>

Show me the contents of /db/myapp/config/settings.xml.

Delete the resource at /db/myapp/tmp/old-import.xml.

Combine the two

Look up xmldb:store, then use it to save
  <todo><item>Write tests</item></todo>
as /db/scratch/todo.xml.

Working with TEI corpora

If your collections contain TEI/XML, you can drive most of the day-to-day philological work straight from the chat.

Metadata from <teiHeader>

List the title, author and publication date for every document in
/db/corpus by reading the teiHeader.

Named entities

Extract all <persName> values from /db/corpus, deduplicate them and
return a frequency-ordered list.

Find every occurrence of <placeName ref="..."/> in /db/corpus/letters
and group them by the `ref` attribute.

Provenance and curation

Show me documents in /db/corpus that have been modified in the last 7
days, ordered by modification date.

Validate that every TEI file in /db/corpus has a non-empty
<sourceDesc>; list the offending paths.

Lightweight stylometry / text analysis

Count the occurrences of <said> per speaker across /db/corpus/plays
using the Lucene full-text index for speed.

Importing and exporting

Store this TEI fragment as /db/corpus/drafts/letter-042.xml:
  <TEI xmlns="http://www.tei-c.org/ns/1.0"> ... </TEI>

Apply /db/corpus/xslt/tei-to-html.xsl to
/db/corpus/letters/letter-001.xml and return the HTML.

Tip — for namespace-aware queries (every TEI document lives under xmlns="http://www.tei-c.org/ns/1.0"), ask Claude to declare declare default element namespace "http://www.tei-c.org/ns/1.0"; at the top of the XQuery. The model is happy to do this when you mention "TEI" in the prompt.

Ecosystem

A few projects you'll likely want next to this one:

• TEI Publisher — the most common front-end for TEI editions on top of eXist-db

• TEI Guidelines — the canonical reference for the encoding scheme

• eXide — the in-browser IDE bundled with eXist-db; useful when Claude hands you a query you want to tweak by hand

Testing

The project ships with unit tests for the offline pieces (documentation lookup and config resolution). They use Node's built-in test runner, so there's no extra dependency:

npm test

Tests inject a small fixture cache directly, so you don't need to run the scraper or have an eXist-db instance up.

For end-to-end checks against a real database, the easiest path is:

# 1. Start eXist-db (Docker is the quickest)
docker run -d --rm -p 8080:8080 --name exist existdb/existdb:latest

# 2. Wait until http://localhost:8080/exist returns 200, then:
EXISTDB_PASSWORD=  node dist/index.js
# and exercise the tools through your MCP client

Troubleshooting

fundocs.json not found at …/dist/cache/fundocs.json The build step couldn't find the documentation cache. Run npm run scrape (or npm run scrape -- --offline for the fixture) and then npm run build again — the post-build step copies the cache into dist/.

Connection error: Cannot reach eXist-db at … The query/collection tools couldn't open a TCP connection. Check that:

• eXist-db is running and reachable at EXISTDB_ENDPOINT

• the URL has no trailing /rest (the server appends it automatically)

• a firewall isn't blocking the port

Query error (HTTP 401) The credentials are wrong or missing. Set EXISTDB_USER / EXISTDB_PASSWORD, or pass username / password explicitly in the tool call. The default admin user has an empty password on a fresh install.

XQuery syntax error at line N, column M This is the eXist-db parser talking — the query has invalid XQuery. Use existdb_validate_xquery to iterate quickly without executing.

The server starts but Claude doesn't see the tools

• Make sure the args path in your MCP config points to the compiled dist/index.js, not the TypeScript source

• Restart Claude Code / Claude Desktop after editing the MCP config

• Check Claude's MCP logs for stderr from the server

Regenerating the documentation cache

The cache is tied to the eXist-db version it was scraped from. Regenerate it whenever you upgrade or want fresher docs:

npm run scrape   # ~2–3 minutes, polite 300 ms delay between namespaces
npm run build

Project structure

existdb-mcp/
├── src/
│   ├── index.ts                  # MCP server entry point
│   ├── types/
│   │   └── index.ts              # Shared TypeScript types
│   ├── tools/
│   │   ├── docs.ts               # Documentation lookup tools
│   │   ├── query.ts              # XQuery execution tools
│   │   ├── query-helpers.ts      # Shared auth / config helpers
│   │   └── collections.ts        # Collection management tools
│   └── cache/
│       └── fundocs.json          # Generated by `npm run scrape`
├── scripts/
│   ├── scrape-fundocs.ts         # Fundocs scraper
│   └── copy-cache.mjs            # Post-build cache copy
├── test/
│   ├── fixtures.ts               # Deterministic fundocs fixture
│   ├── docs.test.ts              # Documentation tools tests
│   └── query-helpers.test.ts     # Config / auth helpers tests
├── dist/                         # Compiled output (npm run build)
├── package.json
├── tsconfig.json
└── README.md

Contributing

Issues and pull requests are welcome. Please run npm test and npm run build before submitting.

License

MIT — see LICENSE.