pretalx-mcp (local-only)

pretalx-mcp is a local MCP server for pretalx. It is designed to run as a child process of a local MCP client and communicates over stdio only.

Local MCP client -> local stdio MCP process -> outbound HTTPS -> pretalx API

Only outbound HTTPS requests to pretalx are expected. No inbound network service is exposed.

Features

• Local-only architecture (no hosted or remote MCP endpoint).
• stdio is the only supported transport.
• Read-only by default (PRETALX_READ_ONLY=true).
• Environment-based configuration via .env.
• MCP tools for events, submissions, speakers, schedule, and optional reviews.

Installation

1. Install Python 3.11+ and uv.
2. Clone this repository.
3. Sync dependencies:

uv sync

Configuration

Copy .env.example to .env and fill in local values:

cp .env.example .env

Variable	Required	Default	Notes
PRETALX_BASE_URL	Yes	none	pretalx base URL, no trailing slash
PRETALX_API_TOKEN	Yes	empty	use a local token; never commit
PRETALX_DEFAULT_EVENT	No	empty	optional fallback event slug
PRETALX_TIMEOUT_SECONDS	No	20	HTTP timeout in seconds
PRETALX_VERIFY_TLS	No	true	keep TLS verification enabled
PRETALX_READ_ONLY	No	true	must remain true for v1
PRETALX_TRANSPORT	No	stdio	only supported value is stdio

Running locally

Run the server locally with stdio transport:

uv run pretalx-mcp

This process is intended to be launched by an MCP client, not exposed as a web service.

MCP client configuration

See docs/MCP_CLIENT_SETUP.md for a generic MCP client JSON configuration using:

uv --directory /path/to/pretalx-mcp run pretalx-mcp

Security notes

• Keep API tokens local and out of git history.
• Prefer least-privilege, read-only pretalx tokens.
• Never log or print tokens.
• Do not run this project as a hosted/cloud MCP service.

See docs/SECURITY.md for full guidance.

Development

Common local commands:

uv sync
uv run ruff check .
uv run pytest

Project docs and metadata files:

• .env.example
• .gitignore
• docs/SECURITY.md
• docs/MCP_CLIENT_SETUP.md

Roadmap

• Deliver read-only organiser workflows first.
• Keep transport local (stdio) and avoid any remote server mode.
• Add optional review-oriented tools where API permissions allow.
• Consider future write tools only behind explicit safety checks and opt-in configuration.