Luma Events MCP Server

A FastMCP server that discovers events from Luma — combining the Discover feed and subscribed calendars — with distance filtering and ICS export. No API key required for basic discovery.

How it works

Luma's Discover API has two endpoints that behave very differently:

• Category search (e.g. AI, Tech, Food) — returns hundreds of events with rich tagging, but only for your home region. Great depth, geographically locked.
• City search (e.g. Paris, London, Tokyo) — returns a curated set of ~20–40 top/featured events for that city. Broad coverage, smaller set.

This MCP uses both via two search modes:

• Home mode (default, no city param) — searches your preferred categories via the Category API. Deep, rich results filtered by address and distance.
• Travel mode (pass a city) — fetches the curated top events for that city via the Place API.

On first run, the server returns popular events near you (geo-biased by IP), then walks you through setting up categories, address, and login for progressively richer results.

Tools

Setup

Prerequisites

• Python 3.10+
• uv (recommended) or pip

Install

git clone <this-repo>
cd "Luma Cal MCP"
uv venv .venv --python 3.12
source .venv/bin/activate
uv pip install -e .

Subscribed calendars (optional)

To access events from calendars you follow on Luma, install the optional auth dependencies:

uv pip install -e ".[auth]"
playwright install chromium

First run

On first use, the raw Discover feed returns hundreds of popular events near you (geo-biased by IP). The server then walks you through setup one prompt at a time to narrow results:

1. Address — asks for your location and preferred search radius, which dramatically reduces the result set to events near you.
2. Categories — asks which topics interest you (from: tech, ai, food, arts, climate, fitness, wellness, crypto) for focused discovery.
3. Login — asks whether to log in for subscribed calendars.

Each prompt appears after returning results, so you see events immediately. After you configure a preference, the search reruns automatically and the next prompt appears. You can respond "not now" (prompt reappears next time) or "never" (permanently dismissed).

Configure

Use set_preferences to save defaults that persist across restarts:

set_preferences(address="3180 18th St, San Francisco", max_distance_miles=15)
set_preferences(categories=["ai", "tech"])

Run

# stdio transport (for Cursor, Claude Desktop, etc.)
fastmcp run src/luma_mcp/server.py

# or directly
python -m luma_mcp.server

Authentication

Subscribed calendars require a Luma session cookie. The server handles this automatically via an inline login flow.

How it works:

1. First call — after results, the server prompts for login. The agent asks you in chat.
2. Login — the agent calls search_events with login=true. A Chromium browser opens to lu.ma/signin; log in normally. The session cookie is stored in the local SQLite DB.
3. Decline — the agent calls search_events with skip_login_days=N to defer (0 = ask next time, -1 = never).
4. Returning user, cookie expired — the browser opens automatically for re-authentication.
5. Validation — the stored cookie is validated against Luma's API

... View full README on GitHub