foundry-rest-api-mcp-server
An MCP server that wraps the ThreeHats Foundry VTT REST API relay so an LLM agent can author and manage Foundry VTT content, primarily creating D&D 5e creatures, through native MCP tools.
README
foundry-rest-api-mcp-server
An MCP (Model Context Protocol) server that wraps the ThreeHats Foundry VTT REST API relay, so an LLM agent (Claude Code / Claude Desktop) can author and manage Foundry VTT content — primarily creating D&D 5e creatures — through native MCP tools instead of raw HTTP calls.
Status: V1 implemented. The stdio MCP server lives under
src/with unit tests undertest/. See Install & configure below to run it. The original plan that drove the build is inIMPLEMENTATION_PLAN.md.
What this is
The relay (github.com/ThreeHats/foundryvtt-rest-api-relay, public host https://foundryrestapi.com) bridges a running Foundry world to a REST API over a WebSocket. A Foundry-side module connects out to the relay; external clients then call REST endpoints, authenticating with an x-api-key header and targeting a world via a clientId query param.
This MCP server is a thin, typed client of that relay. It does not talk to Foundry directly and requires no Foundry-side code of its own (the relay's module handles that). That is the key differentiator from existing Foundry MCP servers, which all ship a custom Foundry module + socket (see IMPLEMENTATION_PLAN.md § Prior Art).
v1 scope (locked)
v1 implements the relay's document-authoring surface — these five scopes only:
| Scope | Endpoints | Capability |
|---|---|---|
clients:read |
GET /clients |
Discover connected worlds → pick clientId |
search |
GET /search |
Find actors/items/etc. by name (world + compendiums) |
entity:read |
GET /get |
Read any document (by UUID or current selection) |
entity:write |
POST /create, PUT /update, DELETE /delete, /give, /remove, /increase, /decrease, /kill |
Create / update / delete documents; tweak actor inventory & attributes |
structure:write |
POST /create-folder, DELETE /delete-folder |
Organize content into folders |
This delivers the headline use case end-to-end: build a dnd5e NPC/monster (abilities, HP, AC, CR, traits, embedded attacks) as a real Foundry actor, search/read existing creatures to clone-and-tweak, and file everything into folders.
Everything else in the relay (dice rolls, chat, encounters/combat, scenes, canvas/tokens, effects, files, users, macro execution, streaming events) is out of v1 scope and documented as clearly-marked Future Phases — each unlocks by adding one scope and one tool module.
Optional companion: adding
structure:read(GET /structure,GET /get-folder) would let us find existing folders, not just create them. Cheap and read-only; deferred for now.
Architecture decisions (the short list)
- Language/runtime: TypeScript, Node ≥ 18 (target Node 20 LTS), ESM-only.
- SDK:
@modelcontextprotocol/sdk@^1.29(the shipped 1.x line). The renamed V2 SDK (@modelcontextprotocol/server) is not stable until ~mid-2026; migration later is mechanical. Build on 1.x now. - Transport: stdio.
- HTTP: native
fetch(no axios). - Validation:
zod(peer of the SDK). - Build/package:
tsup→ singlenpx-runnable bin. - Config: env vars —
FOUNDRY_API_KEY(thex-api-key),FOUNDRY_RELAY_URL(defaulthttps://foundryrestapi.com),FOUNDRY_CLIENT_ID(optional; auto-resolves if one world is online),FOUNDRY_USER_ID(optional; omit = GM-level). - Tool surface: small set of mostly single-purpose tools (~9 in v1), façades only where operations are homogeneous. Per-endpoint-explosion is explicitly avoided — LLM tool-selection reliability degrades past ~30–40 tools and every tool is permanent context cost.
Install & configure
Build from source
npm install
npm run build # → dist/index.js (executable, #!/usr/bin/env node)
npm test # unit tests (mocked fetch)
npm run inspect # launch the MCP Inspector against dist/index.js
Environment variables
| Var | Required | Default | Purpose |
|---|---|---|---|
FOUNDRY_API_KEY |
yes | — | Sent as the x-api-key header. Must include all five v1 scopes (see below). |
FOUNDRY_RELAY_URL |
no | https://foundryrestapi.com |
Relay base URL. Set to http://localhost:3010 for a self-hosted relay. |
FOUNDRY_CLIENT_ID |
no | auto-resolves | Which connected world to target. Auto-resolves when exactly one world is online. |
FOUNDRY_USER_ID |
no | GM-level | Scope actions to a Foundry user's permissions. Omit for GM access. |
On startup the server logs the online world(s) to stderr and disables foundry_create_creature if the active world's system isn't dnd5e.
Tools (v1)
foundry_list_worlds · foundry_search · foundry_get_entity · foundry_create_entity · foundry_update_entity · foundry_delete_entity · foundry_modify_actor · foundry_create_creature · foundry_manage_folder
Claude Code (.mcp.json)
Add to .mcp.json in your project root (or via claude mcp add):
{
"mcpServers": {
"foundry": {
"command": "node",
"args": ["/absolute/path/to/foundry-rest-api-mcp-server/dist/index.js"],
"env": {
"FOUNDRY_API_KEY": "your-scoped-key",
"FOUNDRY_CLIENT_ID": "fvtt_8bfa06d76c0c1ac5"
}
}
}
}
Claude Desktop (claude_desktop_config.json)
~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):
{
"mcpServers": {
"foundry": {
"command": "node",
"args": ["/absolute/path/to/foundry-rest-api-mcp-server/dist/index.js"],
"env": {
"FOUNDRY_API_KEY": "your-scoped-key",
"FOUNDRY_CLIENT_ID": "fvtt_8bfa06d76c0c1ac5"
}
}
}
}
Once published to npm, replace
"command": "node", "args": ["…/dist/index.js"]with"command": "npx", "args": ["-y", "foundry-rest-api-mcp-server"].
Live smoke test (optional)
A gated end-to-end test creates then deletes a throwaway JournalEntry against the real relay:
FOUNDRY_LIVE_TEST=1 FOUNDRY_API_KEY=... FOUNDRY_CLIENT_ID=... npm test
How to use this repo (for the implementing agent)
- Read
IMPLEMENTATION_PLAN.mdtop to bottom. Work the phases in order; each step lists its goal, the files it touches, the exact work, and an acceptance check. - Read
EXECUTION_STRATEGY.mdfor how to run the build — recommended model/effort per ticket, the sequential-foundation → parallel-fan-out → sequential-convergence order, and the branch-per-work-stream (trunk-based) strategy with worktrees + CI gates. - Use
API_SPEC.mdas the authoritative endpoint reference (request/response shapes, scopes, quirks, the dnd5e NPC schema notes). - Pull granular, ordered tasks from the V1 project board / issues (milestone
V1 — Content Authoring). Each issue carries goal, endpoint+scope, file paths, acceptance criteria, andDepends on #N. - The
reference/folder holds the relay's own docs/examples, mirrored verbatim from the upstream MIT-licensed repo for offline ground-truth (attribution inreference/SOURCE.md; regenerate withscripts/fetch-reference.sh). Filenames mirror their repo paths (e.g.docs_md_api_entity.md,docs_examples_dnd5e-examples.json).
Decision records & specs
decisions/— architecture & management decisions as ADRs, numberedNNNN-*.md. Each captures one decision (status · context · decision · consequences) and is immutable once Accepted; a later ADR supersedes an earlier one by reference, so history is never edited in place. Index + convention:decisions/README.md.specs/— technical component specs, numberedNNNN-*.md— the internal contracts the code implements (relay client, tool surface, dnd5e NPC builder). Index:specs/README.md.
The top-level docs (IMPLEMENTATION_PLAN.md, EXECUTION_STRATEGY.md, API_SPEC.md) are the elaborated references; the ADRs are the atomic, supersede-able record of why.
Getting a properly-scoped API key (do this before any write works)
A read-only key returns 403 API key lacks required scope: entity:write. You cannot add scopes to an existing key value — mint a new one that includes all five v1 scopes:
- Dashboard: log in at
https://foundryrestapi.com→ API Keys → Create Scoped Key → tickclients:read,search,entity:read,entity:write,structure:write→ copy the key (shown once). - Programmatic (device flow):
POST /auth/key-requestwith{"appName":"foundry-mcp","scopes":["clients:read","search","entity:read","entity:write","structure:write"]}→ open the returnedapprovalUrl→ pollGET /auth/key-request/:code/statusuntilapproved(the response then includesapiKey).
The known target world for this project is "Wilds Beyond the Witchlight", clientId fvtt_8bfa06d76c0c1ac5 (dnd5e 5.2.2, Foundry v13.351).
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.