Two artifacts: the .mcpb bundle provides the tools, the skill zip provides the how-to knowledge. Install both.
- Prerequisite: none — the bundle uses Claude Desktop's managed
uvruntime. - Download
gds-agent-<version>.mcpbfrom the GitHub releases page. - Double-click the file (or drag it into Claude Desktop → Settings → Extensions) and click Install.
- Fill in the configuration form: Neo4j URI, username, password (stored in the OS keychain), optional database name, and optional Aura API credentials for Aura Graph Analytics session mode.
- Verify: in a new chat, the tools menu should list
gds-agenttools such asget_node_labels.
- Prerequisite: Settings → Capabilities → enable Code execution and file creation (skills require it). On Team/Enterprise an admin must enable skills for the organization first.
- Download
neo4j-graph-data-scientist-skill-<version>.zipfrom the same release. - Settings → Customize → Skills → + → upload the zip.
- Toggle the skill on. Verify by asking a graph question — the response should follow the schema → project → algorithm workflow.
Instead of the .mcpb, you can add the server to claude_desktop_config.json by hand (use the absolute path from which uvx as the command):
{
"mcpServers": {
"neo4j-gds": {
"command": "/opt/homebrew/bin/uvx",
"args": ["gds-agent"],
"env": {
"NEO4J_URI": "neo4j://localhost:7687",
"NEO4J_USERNAME": "neo4j",
"NEO4J_PASSWORD": "<your-password>"
}
}
}
}