MCP server

The Reflex MCP server lets AI assistants (Cursor, Claude Desktop, and other MCP-compatible clients) interact with your Reflex account — reading server state, viewing incidents, and triggering deployments.

Setup

1. Generate an API token in your Reflex dashboard under Settings → API Tokens. Prefer a dedicated, least-privilege token (read-only when you do not need the assistant to trigger deployments).
2. Download the MCP server package from the link in your dashboard under Settings → Integrations → MCP
3. Configure your AI client to use the Reflex MCP server with your API token — store the token in user-level or environment config, never in a committed project file (for example Cursor: user MCP settings / env, not a project .cursor/mcp.json checked into git)

Detailed setup instructions for each supported client are provided in the dashboard after generating your token.

What it can do

• Read server state — View connected servers, their health status, and recent metrics
• View incidents — Browse active and resolved incidents with full timelines
• Trigger deployments — Start deployments from within your AI assistant
• Tail logs — Stream recent log output for debugging

Supported clients

• Cursor — Add Reflex as an MCP server in your project settings
• Claude Desktop — Add Reflex to your Claude Desktop MCP configuration

Configuration examples for each client are available in your dashboard after setup.

Security

• All requests are authenticated with your scoped API token
• The MCP server runs locally on your machine — no additional inbound ports are needed
• Use a dedicated API token for MCP access rather than reusing personal tokens; scope it to the minimum permissions you need (omit deploy rights unless the assistant must trigger deployments)
• Do not commit the API token — keep it out of the repository; if it may have leaked, revoke it under Settings → API Tokens and issue a replacement