FoundryMCP

A lightweight MCP (Model Context Protocol) server for FoundryVTT that communicates directly via WebSockets.

Why This Server?

Unlike other FoundryVTT MCP servers that require:

• Installing a custom module on your Foundry server

• Running a headless browser

This server natively authenticates with FoundryVTT and exchanges WebSocket messages directly using the same protocol as the official Foundry client. This makes it:

• Lightweight - No browser overhead, just direct WebSocket communication

• Zero server-side setup - No modules to install on your Foundry instance

• Secure - Uses the same authentication flow as the official client

Related MCP server: FoundryVTT MCP Server

Compatibility

Works with FoundryVTT v13 and v14. The server detects the Foundry generation automatically (via GET /api/status) and uses the correct session-binding method for that version — v14 changed how the WebSocket session is authenticated, and both are handled transparently with no configuration needed.

Security Recommendation

Create a dedicated FoundryVTT user for each game world you want the MCP server to access. Grant that user only the permissions you want the MCP server to have. This provides:

• Fine-grained access control

• Clear audit trail of MCP actions

• Easy revocation if needed

• Isolation between different games/worlds

Installation

npm install
npm run build

Skills

The ./skills directory contains LLM skills for this project. Currently the only skill there is MAIN.skill, and all users should install it because it includes the FoundryVTT-specific guidance and workflows. We welcome PRs that add skills for specific game systems, FoundryVTT modules, or other targeted use cases.

Configuration

Credentials File

Create a file at config/foundry_credentials.json:

[
  {
    "_id": "my-campaign",
    "hostname": "your-foundry-server.com",
    "userid": "your-user-id",
    "password": "your-password"
  },
  {
    "_id": "test-world",
    "hostname": "test.foundry-server.com",
    "userid": "test-user-id",
    "password": "test-password"
  }
]

Fields:

• _id - A user-defined identifier for this credential entry (used to switch between instances)

• hostname - The domain/IP of your FoundryVTT server

• userid - Your Foundry user document ID (find by inspecting Users in Foundry admin panel)

• password - Your Foundry user password

You can configure multiple Foundry instances and switch between them at runtime using the choose_foundry_instance tool.

Environment Variable

Override the default credentials path by setting:

export FOUNDRY_CREDENTIALS=/path/to/credentials.json

Usage with MCP Clients

Add to your MCP client configuration:

{
  "mcpServers": {
    "foundry": {
      "command": "node",
      "args": ["/path/to/FoundryMCP/build/server.js"]
    }
  }
}

Or if installed globally:

{
  "mcpServers": {
    "foundry": {
      "command": "foundry-mcp"
    }
  }
}

Available Tools

Document Retrieval (List)

These tools retrieve all documents of a given type from the connected world.

Parameters:

• max_length (integer, optional): Maximum response size in bytes. Documents are removed from the response until it fits within this limit.

• requested_fields (string[], optional): Specific fields to include. Supports dot-notation for nested fields (e.g. system.attributes.hp). Always includes _id and name. If omitted, all fields are returned. See Selecting nested fields below.

• where (object, optional): Filter documents by field values. See Filtering with where below.

Document Retrieval (Single)

These tools retrieve a single document by ID or name.

Parameters (at least one required):

• id (string): Document ID

• _id (string): Document ID (alias)

• name (string): Document name

• requested_fields (string[], optional): Specific fields to include. Supports dot-notation for nested fields (e.g. system.attributes.hp). Always includes _id and name. If omitted, all fields are returned. See Selecting nested fields below.

World Metadata

get_world

Get world metadata from FoundryVTT (title, system, version, etc.), excluding document collections like actors or items. Use the get_* document tools for collection data.

Document Manipulation

modify_document

Modify an existing document in FoundryVTT.

Parameters:

• type (string, required): Document type. Valid types:

  • Core: Actor, Item, Scene, JournalEntry, Folder, User, Playlist, Macro, RollTable, Cards, ChatMessage

  • Scene objects: Combat, Combatant, ActiveEffect, Drawing, MeasuredTemplate, Note, Tile, Token, Wall, AmbientLight, AmbientSound

• _id (string, required): The document's unique identifier

• updates (object[], required): Array of update objects with fields to modify

Example - Update actor HP:

{
  "type": "Actor",
  "_id": "abc123",
  "updates": [{ "system": { "attributes": { "hp": { "value": 25 } } } }]
}

create_document

Create new documents in FoundryVTT.

Parameters:

• type (string, required): Document type to create

• data (object[], required): Array of document data objects

Example - Create an item:

{
  "type": "Item",
  "data": [{ "name": "Healing Potion", "type": "consumable" }]
}

delete_document

Permanently delete documents from FoundryVTT. This cannot be undone.

Parameters:

• type (string, required): Document type to delete

• ids (string[], required): Array of document _id values to delete

Example:

{
  "type": "Item",
  "ids": ["vlcf6AI5FaE9qjgJ", "abc123def456"]
}

Instance Management

show_credentials

Show all configured Foundry credentials without revealing passwords. Returns the _id, hostname, userid, item_order (zero-based index), and currently_active status for each credential entry.

Example response:

[
  {
    "_id": "my-campaign",
    "hostname": "your-foundry-server.com",
    "userid": "abc123",
    "item_order": 0,
    "currently_active": true
  },
  {
    "_id": "test-world",
    "hostname": "test.foundry-server.com",
    "userid": "def456",
    "item_order": 1,
    "currently_active": false
  }
]

choose_foundry_instance

Switch to a different Foundry instance. Disconnects from the current instance (if any) and connects to the specified one.

Parameters (at least one required):

• item_order (integer): Zero-based index of the credential in the array

• _id (string): The user-defined identifier for the credential entry

Example - Switch by item_order:

{
  "item_order": 1
}

Example - Switch by _id:

{
  "_id": "test-world"
}

Tips

Understanding Document Structure

Document schemas vary significantly between game systems (D&D 5e, Pathfinder, etc.). Use the get_* tools to inspect existing documents before attempting to modify or create new ones.

Filtering with where

All collection retrieval tools (get_actors, get_items, etc.) support the where parameter for filtering results. The where parameter accepts an object with key-value pairs that documents must match.

How it works:

• Each key-value pair in where is a condition that must be satisfied

• All conditions use AND logic - a document must match ALL conditions to be included

• Values are compared using strict equality

Example - Get actors in a specific folder:

{
  "where": {
    "folder": "abcd1234"
  }
}

Example - Get NPC actors in a specific folder:

{
  "where": {
    "folder": "abcd1234",
    "type": "npc"
  }
}

This returns only actors where folder equals "abcd1234" AND type equals "npc".

Example - Get items of a specific type:

{
  "where": {
    "type": "weapon"
  }
}

Example - Combine with other parameters:

{
  "where": {
    "folder": "abcd1234"
  },
  "requested_fields": ["name", "type", "system.quantity"],
  "max_length": 5000
}

Common filter fields:

• folder - Filter by folder ID

• type - Filter by document subtype (e.g., "npc", "character" for actors; "weapon", "armor" for items)

• ownership - Filter by ownership settings

• Any top-level field on the document can be used as a filter key

Selecting nested fields

requested_fields accepts dot-notation paths, so you can pull a single nested value instead of a whole subtree. Requesting system on a character-heavy game system can return tens of kilobytes; requesting system.attributes.hp returns just what you asked for.

{
  "name": "Bezaliel",
  "requested_fields": ["system.age", "system.attributes_physical.strength", "img"]
}

{
  "system": {
    "age": 52,
    "attributes_physical": { "strength": { "value": 1, "final": 1 } }
  },
  "img": "worlds/my-world/assets/bezaliel.jpg",
  "_id": "jJ10LA0URnIaLu7g",
  "name": "Bezaliel"
}

Specifics worth knowing:

• The result keeps its nesting. A request for system.age comes back as {"system": {"age": 52}}, not as a flat "system.age" key, so the shape matches the document you'd get from an unfiltered request.

• Paths sharing a parent merge. ["system.age", "system.sex"] returns both under a single system object.

• A broader path wins over a narrower one. Requesting both system.attributes_physical and system.attributes_physical.strength returns the full attributes_physical object.

• Literal keys take precedence. If a document genuinely has a key named "system.age", that value is returned as-is and the path is not traversed.

• Missing paths are skipped silently. system.nonexistent.path is omitted from the response rather than erroring, and it does not affect the other requested fields.

• Arrays are opaque. You can request an array (system.progress), but you cannot index into one — system.progress.0.name resolves to nothing.

• where does not support dot-notation. It matches top-level fields only, so requested_fields and where follow different rules in the same request.

Response Size Management

When working with large worlds, use max_length and requested_fields to limit response sizes:

{
  "max_length": 10000,
  "requested_fields": ["name", "type", "system.attributes.hp"]
}

max_length applies to list tools only, and drops whole documents from the end of the response until it fits. Narrowing requested_fields shrinks every document instead, so the two combine well.

How It Works

1. Version detection: Queries /api/status to detect the Foundry generation (v13 vs v14), which determines how the session binds to the socket

2. Authentication: Authenticates with FoundryVTT using the same HTTP POST flow as the official client

3. WebSocket Connection: Establishes a persistent Socket.IO/Engine.IO connection, binding the session the way that version expects (query parameter on v13, session cookie on v14), then waits for Foundry to confirm the bound session before issuing requests

4. Message Exchange: Exchanges JSON messages using Foundry's native protocol, replying to Engine.IO keepalive pings to hold the connection open

5. Automatic Reconnection: Handles connection drops and re-authenticates as needed

License

MIT