Installation

You need an Autotask API user: Admin → Resources, license type API User, with an API-tracking integration code and a security level that allows what your tools do. The server finds your API zone from the username.

Claude Code

claude mcp add psarelay -e AUTOTASK_USERNAME=api-user@example.com -e AUTOTASK_SECRET=your-secret -e AUTOTASK_INTEGRATION_CODE=your-code -- npx -y github:GDSTechnology/psarelay

Claude Desktop

Build the bundle from source, then open the .mcpb file it creates in the repo root and enter your username, secret and integration code:

git clone https://github.com/GDSTechnology/psarelay.git && cd psarelay
npm ci && npm run build && npm run pack:mcpb

Docker (HTTP)

For ChatGPT, n8n and other remote clients:

docker run -d --name psarelay -p 8080:8080 -e AUTOTASK_USERNAME=api-user@example.com -e AUTOTASK_SECRET=your-secret -e AUTOTASK_INTEGRATION_CODE=your-code --restart unless-stopped ghcr.io/gdstechnology/psarelay:latest

Clients connect to http://<host>:8080/mcp (MCP Streamable HTTP). GET /health returns the version, API usage and backpressure.

From source

npm ci && npm run build
node dist/entry.js                       # stdio
MCP_TRANSPORT=http node dist/index.js    # HTTP

Production

Releases are cut on every merge to main and published to ghcr.io/gdstechnology/psarelay. In production the server runs as a docker-compose service next to Postgres; the deploy runbook covers layout, environment, verification and rollback.

Updating is one command on the host. Install it once from the image:

docker run --rm --entrypoint cat ghcr.io/gdstechnology/psarelay:latest /app/deploy/update-psarelay.sh > /usr/local/bin/update-psarelay && chmod +x /usr/local/bin/update-psarelay

update-psarelay

It updates only PSArelay service: pulls, keeps the previous image as :rollback, runs database migrations, restarts, waits for /health and prints the version. It refuses while a bulk job is preparing, and refreshes itself from each new image.

Then add the Postgres layer and the admin console.