luma-mcp

lu.ma events MCP server. Create events, set ticket prices, issue coupon codes, list RSVPs, and email attendees. All writes go through draft+confirm so nothing posts to Luma until you explicitly approve. Built on the official Luma public API.

What it does

• Create + update lu.ma events (name, date, timezone, capacity, meeting URL, visibility, cover image)
• Set ticket types and pricing
• Issue coupon codes (calendar-scoped or event-scoped, percentage or fixed-cents discounts, redemption caps)
• List event guests / RSVPs
• Send markdown emails to attendees (pre-work, day-of, follow-up)
• Audit log of every API call (JSONL with key/token redaction)
• Prompt-injection scrubber on user-supplied free-text fields

Tools (v0.1.0)

Reads (call API directly):

• luma_health — auth check + draft store size
• luma_get_event — fetch one event by ID
• luma_list_events — paginated calendar listing
• luma_list_event_guests — RSVPs filtered by status
• luma_list_coupons — calendar or event-scoped

Writes (draft+confirm pairs):

• luma_create_event_draft / luma_create_event_confirm
• luma_update_event_draft / luma_update_event_confirm
• luma_create_ticket_type_draft / luma_create_ticket_type_confirm
• luma_create_coupon_draft / luma_create_coupon_confirm
• luma_send_event_email_draft / luma_send_event_email_confirm

Plus:

• luma_cancel_draft — cancel an unconfirmed draft (idempotent)

14 tools total.

Architecture

luma-mcp/
├── pyproject.toml
├── README.md
├── SETUP.md
├── LICENSE
├── .env.example
├── luma_mcp/
│   ├── __init__.py
│   ├── server.py             FastMCP entry point + .env loader
│   ├── client.py             HTTP wrapper around api.lu.ma/public/v1
│   ├── drafts.py             Thread-safe draft store with TTL
│   ├── audit.py              JSONL audit log with redaction
│   ├── scrubber.py           Prompt-injection neutralizer
│   ├── _shared/              Day-1 shared utilities
│   │   ├── normalize.py      Timezone alias + datetime helpers
│   │   └── types.py          Permissive typed shapes for API responses
│   └── tools/
│       ├── health.py         luma_health, luma_cancel_draft
│       ├── events.py         create / update / get / list events
│       ├── tickets.py        create ticket types
│       ├── coupons.py        create / list coupons
│       ├── guests.py         list event guests
│       └── emails.py         send email to attendees
└── tests/integration/
    └── test_full_pipeline.py 10-step bare-runnable verification

Safety patterns

1. Draft + confirm on every write. The draft tool returns a draft_id; nothing hits Luma until the matching confirm tool is called. Drafts expire after 1 hour and can only be confirmed once.
2. Audit log. Every API call is logged JSONL with ts, tool, action, args (redacted), status, latency_ms. Path: ~/.claude/luma-mcp/audit.log.
3. Token redaction. Strings matching long-key heuristics, or values under keys like api_key, token, cookie, secret, are masked before logging.
4. Prompt-injection scrubber. User-supplied free-text fields (event name, description, email subject, email body) are passed through unicodedata.normalize("NFKC"), stripped of zero-width characters, and have role-spoof prefixes / fake fences wrapped in backticks (preserving operator visibility, neutralizing model effect on downstream reads).
5. Zero LLM in the tool path. Every routing decision is rule-based.

Use cases

• Workshops + paid events — set sticker price, issue community-specific 100%-off coupons, pull guest list, blast pre-work email.
• Multi-event programs — one cohorte announcement, sub-events on Luma, pull RSVPs across sessions.
• Recurring meetups — duplicate event template via update_event_draft, change date, ship.

Install

Open Claude Code, paste:

/plugin marketplace add adelaidasofia/luma-mcp
/plugin install luma-mcp@luma-mcp

Two prerequisites for the server to authenticate:

1. Luma Plus subscription on the calendar you'll use (the public API is gated behind Plus).
2. API key generated from https://lu.ma/settings/api, dropped into ~/.claude/luma-mcp/.env as LUMA_API_KEY=....

After both steps, restart Claude Code. Tools appear under mcp__luma__*.

Verification

cd ~/.claude/luma-mcp && python3 tests/integration/test_full_pipeline.py
# expected: PASSED — luma-mcp v0.1.0 integration test (10 steps green)

Then in a Claude Code session: mcp__luma__luma_health() → {ok: true, user: {...}}.

Rate limit

Luma's public API caps at 200 requests / minute / calendar. The client does not auto-retry on 429; it surfaces the error and lets the operator pace.

Related MCPs

Same author, same architecture pattern (FastMCP, draft+confirm on writes, vault auto-export where applicable):

• slack-mcp — multi-workspace Slack
• imessage-mcp — macOS iMessage
• whatsapp-mcp — WhatsApp via whatsmeow
• apollo-mcp — Apollo.io CRM + sequences
• google-workspace-mcp — Gmail / Calendar / Drive / Docs / Sheets
• substack-mcp — Substack writing + analytics
• parse-mcp — markitdown / Docling / LlamaParse router

Telemetry

This plugin sends a single anonymous install signal to myceliumai.co the first time it loads in a Claude Code session on a given machine.

What is sent:

• Plugin name (e.g. slack-mcp)
• Plugin version (e.g. 0.1.0)

What is NOT sent:

• No user identifiers, names, emails, tokens, or API keys
• No file paths, message content, or anything from your work
• No IP address is stored after dedup processing

Why: Helps the maintainer know which plugins people actually install, so attention goes to the ones that get used.

Opt out: Set the environment variable MYCELIUM_NO_PING=1 before launching Claude Code. The hook will skip the network call entirely. Already-pinged installs leave a sentinel at ~/.mycelium/onboarded-<plugin> — delete it if you want to reset state.

License

MIT. See LICENSE.