Deploy and Host Open Brain on Railway

Open Brain is a per-user AI memory server over MCP (Model Context Protocol). Each team member gets a personal token URL; their memories are fully isolated and follow them across Claude Desktop, claude.ai, Claude Code, and Cursor.

About Hosting Open Brain

One Node.js MCP service plus one Postgres database. The service auto-creates its schema on boot, exposes a streamable-HTTP MCP endpoint per user token, and an admin API for minting users. Search is Postgres full-text - no embedding API keys required. Source: https://github.com/tswic-highdesert/openbrain-railway

Why Deploy Open Brain?

Give a whole team persistent AI memory without a SaaS subscription. Memories are private per user, stored in your own Postgres, and tokens are stored hashed.

Common Use Cases

Team knowledge that follows each person across AI tools
Persistent project context between Claude/Cursor sessions
A self-hosted alternative to per-app AI memory silos

Dependencies for Open Brain Hosting

Postgres (included in this template)

Deployment Dependencies

An MCP-capable AI client (Claude Desktop, claude.ai, Claude Code, Cursor)

Setup after deploy

On the openbrain service set ADMIN_TOKEN to a long random string if empty (openssl rand -hex 24).
If prompted for Postgres variables use: PGDATA=/var/lib/postgresql/data/pgdata, PGPORT=5432, POSTGRES_DB=railway, POSTGRES_USER=postgres, SSL_CERT_DAYS=820, RAILWAY_DEPLOYMENT_DRAINING_SECONDS=60.
Create users: POST /admin/users with Authorization: Bearer and body {"name": "Jane"}.
Send each person their returned MCP URL - they paste it into their AI tool as a custom connector.