uefnmcp-connector

A local MCP server that grounds your AI assistant in the hosted Verse Cortex knowledge base — Epic's full UEFN documentation plus the Verse API index — through the metered verseisland.com gateway.

Plug it into Claude Desktop, Cursor, or Claude Code with your API key and your AI can now answer UEFN / Verse questions from real, current Epic docs and the indexed Verse API instead of guessing from stale training data. The KB grounding is the whole point: local models are routinely wrong about Verse syntax and API names — these tools fix that.

Tools it exposes

Each call goes through the metered gateway: it authenticates your key, enforces your tier's daily limits, meters usage, and returns ranked KB results. The connector never touches the vector DB or the embedding server directly.

Install

pip install uefnmcp-connector

or run it without installing, straight from PyPI, via uvx:

uvx uefnmcp-connector            # one-off run, no install

Get an API key

1. Sign in at https://verseisland.com (Discord or email — free during beta).
2. Go to https://verseisland.com/me/api-keys.
3. Click Create key, copy the uefn_sk_... value — it's shown once.

Configure your MCP client

The connector reads two environment variables:

Claude Code (CLI)

One paste:

claude mcp add uefnmcp \
  --env VERSEISLAND_API_KEY=uefn_sk_your_key_here \
  -- uefnmcp-connector

Claude Desktop

Edit claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/, Windows: %APPDATA%\Claude\):

{
  "mcpServers": {
    "uefnmcp": {
      "command": "uefnmcp-connector",
      "env": {
        "VERSEISLAND_API_KEY": "uefn_sk_your_key_here"
      }
    }
  }
}

If uefnmcp-connector isn't on your PATH, use the module form: "command": "python", "args": ["-m", "uefnmcp_connector"] (or "command": "uvx", "args": ["uefnmcp-connector"]).

Cursor

Add to ~/.cursor/mcp.json (or .cursor/mcp.json in a project):

{
  "mcpServers": {
    "uefnmcp": {
      "command": "uefnmcp-connector",
      "env": {
        "VERSEISLAND_API_KEY": "uefn_sk_your_key_here"
      }
    }
  }
}

Restart the client. Ask it something like "search the Verse KB for how to spawn a prop" and it will call verse_kb_search.

Tiers

Free during the beta-launch window (all tiers unlocked, no card). Beta signups lock the founder rate for 12 months.

verse_health is always unlimited. Premium compile/build credit tools (compile_verse, compile_and_fix, …) land in a later release behind the same gateway. Manage keys and see usage at https://verseisland.com/me/api-keys.

Composing with uefn-mcp (local UEFN drive)

This connector is the KB-grounding half. To also drive UEFN — compile Verse, build, place actors on your own machine — pair it with a local UEFN driver server (uefn-mcp, distributed separately), which talks to your local UEFN build setup over Node. Add both servers to your client:

{
  "mcpServers": {
    "uefnmcp": {
      "command": "uefnmcp-connector",
      "env": { "VERSEISLAND_API_KEY": "uefn_sk_..." }
    },
    "uefn-mcp": {
      "command": "node",
      "args": ["/path/to/uefn-mcp/dist/index.js"]
    }
  }
}

Your AI then grounds its Verse in the hosted KB (verse_* tools here) and compiles/runs it locally (uefn-mcp tools) — the full write → ground → compile loop, with the cloud KB doing the heavy lifting and your machine doing the build. (Hosted compile credits via the gateway are on the roadmap; until then, compile stays local via uefn-mcp.)

Development / local testing

pip install -e .
# Point at a local gateway instead of prod:
VERSEISLAND_GATEWAY_URL=http://localhost:3101/mcp-api \
VERSEISLAND_API_KEY=uefn_sk_... \
  uefnmcp-connector            # serves MCP over stdio

A standalone smoke test (no MCP client needed) lives in smoke_test.py.

License

Apache-2.0.