io.github.PhilflowIO/dav-mcp

dav-mcp

Give your AI agents the power of organization — Transform them into orchestrating assistants managing calendars, contacts, and tasks.

Built on 27 production-ready tools spanning CalDAV, CardDAV, and VTODO protocols.

Quick Start

One-click install (Claude Desktop)

1. Download dav-mcp-<version>.mcpb from the latest release. Releases from 4.1.2 on carry this file.
2. Open it with Claude Desktop (macOS or Windows). Claude shows an install dialog.
3. Enter your server URL, username and password. The password is stored as a secret. Leave the authentication method at Basic unless your server needs Digest.

The .mcpb file is an MCP Bundle: the server with its dependencies, so neither Node.js nor npx needs to be installed.

Claude Desktop / Cursor (Local)

Add to your MCP config file:

{
  "mcpServers": {
    "dav-mcp": {
      "command": "npx",
      "args": ["-y", "dav-mcp"],
      "env": {
        "CALDAV_SERVER_URL": "https://dav.example.com",
        "CALDAV_USERNAME": "your_username",
        "CALDAV_PASSWORD": "your_password"
      }
    }
  }
}

Config file locations:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json
• Linux: ~/.config/Claude/claude_desktop_config.json

Restart Claude Desktop after adding the configuration.

n8n (Remote HTTP)

Start the HTTP server:

CALDAV_SERVER_URL=https://dav.example.com \
CALDAV_USERNAME=your_username \
CALDAV_PASSWORD=your_password \
BEARER_TOKEN=your-secret-token \
npx dav-mcp --http

Then in n8n:

1. Add AI Agent node
2. Add MCP Client Tool node and connect to AI Agent
3. Configure:
   • MCP Endpoint: http://localhost:3000/mcp
   • Authentication: Bearer
   • Token: your-secret-token

Custom port:

npx dav-mcp --http --port=8080

Docker

Pull the published image (distroless, runs as non-root, no shell):

docker run -d --name dav-mcp -p 3000:3000 --env-file .env \
  ghcr.io/philflowio/dav-mcp:latest

Tags: latest, the exact release (5.0.0) and the minor line (5.0). Set PORT to serve on a different port; the healthcheck follows it.

Or build from source:

git clone https://github.com/PhilflowIO/dav-mcp.git
cd dav-mcp
cp .env.example .env
# Edit .env with your credentials
docker-compose up

The Orchestration

When partial tools force your AI to improvise, complete tools let it execute precise operations across all components.

Available Tools (27 Total)

CalDAV Tools (12 tools)

1. list_calendars - List all available calendars
2. list_events - List ALL events (use calendar_query for filtered searches)
3. create_event - Create a new calendar event
4. update_event - PREFERRED: Update any event field (SUMMARY, LOCATION, STATUS, custom X-* properties); move or convert an event with start_date/end_date/all_day
5. update_event_raw - Update event with raw iCal data (advanced)
6. delete_event - Delete an event permanently
7. calendar_query - PREFERRED: Search and filter events efficiently by text, date range, or location
8. make_calendar - Create a new calendar collection, optionally with a time zone (an IANA name such as Europe/Berlin, sent as a VTIMEZONE)
9. update_calendar - Update calendar properties (display name, description, color, time zone)
10. delete_calendar - Delete a calendar and all its events
11. calendar_multi_get - Batch fetch multiple specific events by URLs
12. freebusy_query - Find free and busy time in a range ("when am I free?"), calculated client-side

CardDAV Tools (8 tools)

1. list_addressbooks - List all available address books
2. list_contacts - List ALL contacts (use addressbook_query for filtered searches)
3. create_contact - Create a new contact (vCard)
4. update_contact - PREFERRED: Update any contact field (FN, EMAIL, TEL, ORG, ADR, custom X-* properties)
5. update_contact_raw - Update contact with raw vCard data (advanced)
6. delete_contact - Delete a contact permanently
7. addressbook_query - PREFERRED: Search and filter contacts efficiently by name, email, or organization
8. addressbook_multi_get - Batch fetch multiple specific contacts by URLs

VTODO Tools (7 tools)

1. list_todos - List ALL todos/tasks (use todo_query for filtered searches)
2. create_todo - Create a new todo/task with optional due date, priority, status
3. update_todo - PREFERRED: Update any todo field (SUMMARY, STATUS, PRIORITY, DUE, PERCENT-COMPLETE, custom X-* properties)
4. update_todo_raw - Update todo with raw VTODO iCal data (advanced)
5. delete_todo - Delete a todo/task permanently
6. todo_query - PREFERRED: Search and filter todos efficiently by status/due date
7. todo_multi_get - Batch fetch multiple specific todos by URLs

Time zones

An event or todo written without a time zone (a "floating" time such as DTSTART:20261010T090000, or an all-day date) is read in its calendar's time zone, as CalDAV specifies (RFC 4791 §9.9): the zone set with make_calendar / update_calendar, which list_calendars shows. A calendar without one is read in the zone dav-mcp runs in: the TZ environment variable (for example TZ=Europe/Berlin in Docker or an HTTP deployment), else the system's. A time without a zone given to the create and update tools is read the same way, and an event stored without a zone stays without one when it is moved. freebusy_query prints its slots in the calendar's zone and names it.

Real-World Applications

n8n Automation Workflows

• Meeting Management: "Show me all Friday meetings" → calendar_query with date filter returns only relevant events
• Contact Search: "Find everyone at Google" → addressbook_query with org filter finds matches efficiently
• Task Reporting: "Show overdue high-priority tasks" → todo_query with filters returns specific results
• Scheduled Cleanup: Daily cron job deletes completed tasks using targeted queries

Claude Desktop Integration

• Quick Event Creation: "Create team meeting tomorrow 2 PM" → create_event executes immediately
• Contact Lookup: "What's Sarah's email?" → addressbook_query with name filter finds contact
• Calendar Overview: "What's on my calendar next week?" → calendar_query with date range shows events
• Calendar Management: "Create a new calendar called Project Luna" → make_calendar creates collection

Works Across All Major Providers

Works with any CalDAV/CardDAV server that follows RFC 4791 and RFC 6352:

• Nextcloud - Full support
• Baikal - Full support
• Radicale - Full support
• iCloud - Works with app-specific password
• Any RFC-compliant server - Standard protocol support

Authentication

AUTH_METHOD selects how dav-mcp logs in (case-insensitive):

dav-mcp needs Node.js 22 or newer. A wrong configuration — an unknown AUTH_METHOD value, or missing credentials for the chosen method — stops the server at startup instead of falling back to Basic.

Google Calendar (OAuth2)

For Google Calendar, use OAuth2 authentication:

{
  "mcpServers": {
    "dav-mcp": {
      "command": "npx",
      "args": ["-y", "dav-mcp"],
      "env": {
        "AUTH_METHOD": "OAuth",
        "GOOGLE_USER": "your@gmail.com",
        "GOOGLE_CLIENT_ID": "your-client-id",
        "GOOGLE_CLIENT_SECRET": "your-client-secret",
        "GOOGLE_REFRESH_TOKEN": "your-refresh-token"
      }
    }
  }
}

Security

• Input Validation: All inputs validated with Zod schemas before execution
• Rate Limiting (HTTP mode): 100 requests per 15 minutes per client address, counted before the bearer token is checked. Clients on loopback or a private network (10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16) — another container on the same Docker network, for example — get 10,000. Limitation: dav-mcp sees the address of whatever connects to it. Behind a Docker port mapping or a reverse proxy that is the bridge gateway or the proxy, so all outside clients share that one address, its one counter and its raised limit. Rate-limit per client at the proxy if you expose the HTTP transport.
• Bearer Auth: Token authentication for HTTP transport
• No Credential Storage: Pass-through only, never logged or cached
• Structured Logging: Audit trail with request IDs, no PII exposure
• Tool-Call Log Off by Default: LOG_TOOL_CALLS=true records every tool call with its arguments, for debugging — to ~/.local/state/dav-mcp/tool-calls.jsonl (or $XDG_STATE_HOME, %LOCALAPPDATA% on Windows, or TOOL_CALL_LOG_FILE), readable by you only. TOOL_CALL_LOG_MODE=console sends it to stderr instead.
• CORS Protection: Whitelist origins, block cross-site attacks

Privacy Policy

dav-mcp runs on your machine (or your own server) and talks only to the DAV server you configure. It has no telemetry and shares nothing with its maintainers or third parties. Tool results go to the AI assistant that called the tool. Credentials and calendar data are not stored by dav-mcp; the optional tool-call log is off by default. Full policy: PRIVACY.md. Contact: hello@philflow.io.

Documentation

• MCP Specification - Model Context Protocol docs
• tsdav Docs - CalDAV/CardDAV library reference
• CalDAV RFC 4791 - CalDAV protocol specification
• CardDAV RFC 6352 - CardDAV protocol specification

Contributing

Pull requests are welcome! Please read CONTRIBUTING.md for guidelines.

License

MIT License - see LICENSE for details

Acknowledgments

Built with:

• tsdav - Excellent TypeScript CalDAV/CardDAV library
• tsdav-utils - Field-agnostic utility layer for RFC-compliant field updates
• MCP SDK - Model Context Protocol by Anthropic
• ical.js - RFC-compliant iCalendar parser

Questions? Issues? Create a GitHub issue

Built for AI agents managing calendars, contacts, and tasks