Curator — Museums Victoria Collections MCP Server

An MCP server that turns the Museums Victoria Collections API (1.15M+ records) into natural-language discovery tools for MCP clients such as Copilot Studio.

Overview

Curator is a Model Context Protocol server exposing the Museums Victoria Collections API as a set of tools. It lets an AI agent search and explore specimens, items, articles, and species across the museum's Natural Sciences and Humanities collections — with taxonomy, images, provenance, and dates.

It runs as a streamable-HTTP MCP service (built with FastMCP) and is served over HTTPS on Azure Container Apps.

Tools

Related MCP server: ALA API MCP Server

Architecture

flowchart LR
    Dev([Developer]) -->|git push main| GH[GitHub Actions]

    subgraph Azure["Azure · rg-curator"]
        direction LR
        ACR[("Container Registry<br/><i>ABAC mode</i>")]
        APP["Container App<br/>HTTPS &middot; always-warm"]
        ACR -->|"pull via<br/>managed identity"| APP
    end

    GH -->|"build &amp; push<br/>(SP AAD token)"| ACR
    APP -->|"serves /mcp"| CS["Copilot Studio<br/>agent"]

    TF[/"Terraform<br/>infra as code"/] -. provisions .-> Azure

Flow: a push to main triggers GitHub Actions, which builds the image and pushes it to the registry; the Container App pulls that image via a managed identity and serves the MCP endpoint over HTTPS, which the Copilot Studio agent calls. Terraform declares the same Azure resources.

• Server: curator_mcp_server.py — FastMCP streamable-HTTP app, MCP endpoint at /mcp on port 8000.

• CI/CD: .github/workflows/deploy-azure.yml — builds the image, pushes to ACR with the workflow service principal's AAD token, and deploys to Container Apps. No registry passwords in the pipeline.

• Auth: the container app pulls images via a user-assigned managed identity; the ACR admin user is disabled. Because the registry runs in ABAC mode, the ABAC-native Container Registry Repository Reader/Writer roles are used (not the classic AcrPull/AcrPush).

• IaC: infra/terraform/ — the whole Azure stack as Terraform (see its README).

Local development

python -m venv venv && source venv/bin/activate
pip install -r requirements.txt

# Run the MCP server on http://localhost:8000/mcp
python curator_mcp_server.py

Or with Docker:

docker compose up --build   # serves on http://localhost:8000/mcp

Deployment

Deployment is automated: pushing to main runs the GitHub Actions workflow, which builds the image and rolls the Container App to the new build. The workflow needs one repository secret, AZURE_CREDENTIALS (a service principal with Contributor on the resource group).

To provision the underlying infrastructure declaratively, use the Terraform in infra/terraform/.

Data source & acknowledgement

This project is built on the Museums Victoria Collections API (https://collections.museumsvictoria.com.au/api) and would not exist without it. All collection data — specimens, items, articles, and species — is provided by Museums Victoria. Huge thanks to Museums Victoria for making this data openly available.

Please note when using the data:

• Licensing is per-record. Individual records carry their own licence (commonly CC BY 4.0, Public Domain Mark, or CC0) — check each record's licence and rightsStatement fields and attribute accordingly.

• A User-Agent header is required on every API request; this server sends one.

• Some records contain culturally sensitive content. Handle Aboriginal and Torres Strait Islander cultural material with appropriate respect.

This project is not affiliated with or endorsed by Museums Victoria.

License

MIT — see the note above; the museum data is licensed per-record by Museums Victoria and is not covered by this repository's licence.