Quickstart
Get OpRelay running in under five minutes. All you need is Docker.
Prerequisites
- Docker and Docker Compose
- macOS with Apple Silicon (recommended) or any Linux server
- An MCP-compatible client (Claude Code, Cursor, Codex, etc.)
1. Clone and start
git clone https://github.com/oprelay/oprelay.git
cd oprelay
docker compose up
This starts:
- OpRelay runtime — broker, MCP server, execution engine
- PostgreSQL — facts, decisions, tasks, runs, audit trail
- Dashboard — operator control panel at
http://localhost:3000
2. Connect your first agent
Add OpRelay to your MCP client config:
mcp.json
{
"mcpServers": {
"oprelay": {
"url": "http://localhost:3000/mcp",
"headers": {
"Authorization": "Bearer your-token"
}
}
}
}
Start a new session in your MCP client. OpRelay tools auto-discover — you'll see 30+ operations available immediately.
3. Store your first fact
Ask your agent to store a fact:
Store a fact in OpRelay: key "stack.language" with value "typescript"
in the "infra" category for project "myproject"
Or call the tool directly:
await oprelay.call("upsert_fact", {
project_key: "myproject",
category: "infra",
key: "stack.language",
value_json: { type: "typescript", version: "5.4" }
});
4. Create a task
await oprelay.call("core_create_task", {
project_key: "myproject",
title: "Set up CI pipeline",
priority: 2,
assigned_to: "worker-01"
});
5. Open the dashboard
Navigate to http://localhost:3000 to see the operator dashboard — timeline view, agent presence, context browser, and audit trail.
Next steps
- Set up the Mac appliance — full stack with Ollama for local inference
- Connect multiple clients — Claude Code + Codex + Cursor
- Learn the primitives — facts, decisions, tasks, runs