Skip to main content

Set up the testing environment

The quickest way to see the workflow end to end is the AI-Assisted Data Discovery demo: a self-contained instance that starts Arranger, the MCP server, and a portal preloaded with a representative sample of the OICR Drug Discovery data. It is locally deployable and accessible to pilot users and developers.

Prerequisites

Install and verify the following before you start.

1. WSL2 Setup (windows only)

Docker Desktop on Windows runs on WSL2, so configure it before installing Docker Desktop below:

  1. Install WSL2
  2. Use Ubuntu or another Linux distribution within WSL2
  3. Enable Docker Desktop's WSL2 integration (Docker Desktop → Settings → Resources → WSL Integration)
  4. Run all commands from a Bash terminal inside WSL2, not PowerShell or Command Prompt. To open one, search for your Linux distribution (e.g. "Ubuntu") in the Start menu.
2. Git installed

Download from git-scm.com if the command is not recognised.

3. Docker Desktop (28.0.0 or later)

Once installed, open Docker Desktop → Settings → Resources and set:

  • CPUs: 4+ cores (8 recommended)
  • Memory: 8 GB minimum
  • Disk: 10 GB+ available

Please ensure docker --version and docker compose version both return version numbers, and Docker Desktop is running with 4+ CPUs and 8 GB+ memory allocated

The local model runtime (LM Studio) is installed later, in Connect a host application.

Start the demo

Clone the demo branch and start the stack:

git clone -b docs-demo/ai-assisted-data-discovery --recurse-submodules https://github.com/overture-stack/prelude.git
cd prelude
make demo

make demo runs system checks, starts the stack via Docker Compose, and loads the sample catalogues. The first run also downloads and builds the container images, so it can take a while depending on your connection. When it completes:

  • the portal is at http://localhost:3000 (the stack does not open a browser for you), and
  • the MCP server is at http://localhost:3100/mcp.

To stop the stack, run make down; to wipe all data and start clean, run make reset.

Windows

Run make demo from a Bash terminal inside WSL2. The MCP demo branch drives the stack through the Makefile, so use WSL2 rather than native PowerShell.

What's running

docker ps should list the stack. All ports bind to 127.0.0.1 only:

ServicePortRole
Stage (portal UI)3000Portal frontend
Arranger5050GraphQL API and introspection endpoints
Arranger MCP3100MCP server, Streamable HTTP at /mcp
OpenSearch9200Search engine
PostgreSQL5435Persistent storage

The demo data

The data is a representative sample drawn from the Drug Discovery Portal, roughly 1,000 rows in each of four gene-statistics catalogues. The upstream portal spans about 405 million records across roughly 20,000 genes and 32 cancer types, and its datasets are related by gene (HUGO symbol), so a cohort filtered in one catalogue can be refined against the others:

CatalogueContentsSample coverage
correlationGene correlation dataDLBC only
mutationGene mutation dataBRCA only
expressionGene expression profiles32 cancer types
proteinProtein interaction statisticsSampled
donorDonor recordsSampled

Verify

Confirm Arranger and the MCP server are up before connecting a host:

# Arranger introspection lists the loaded catalogues
curl http://localhost:5050/introspection

# The MCP container is running and listening on 3100
docker ps | grep arranger-mcp
docker logs arranger-mcp # should show it connected to Arranger and listening

Once the stack is up we can move on to connecting a host application.