MCP Quickstart

Connect any MCP-compatible AI agent to the Latent Space Hub wiki-base.

Setup (2 minutes)

Add this to your Claude Code or Cursor MCP config:

{
  "mcpServers": {
    "latent-space-hub": {
      "command": "npx",
      "args": ["-y", "latent-space-hub-mcp"],
      "env": {
        "TURSO_DATABASE_URL": "your-turso-url",
        "TURSO_AUTH_TOKEN": "your-turso-token"
      }
    }
  }
}

Restart your AI assistant. The ls_* tools should now be available.

What You Can Do

Search the Wiki-Base

Ask your agent:

• "Search Latent Space for episodes about agents"
• "Find what was said about RAG in recent podcasts"
• "Who has appeared on the Latent Space podcast?"

Explore Connections

• "What topics does Andrej Karpathy connect to?"
• "Show me the most connected entities in the graph"
• "Trace the relationship between OpenAI and agents"

Run Custom Queries

• "How many podcast episodes are in the hub?"
• "What are the most recent articles?"
• "List all dimensions with their node counts"

Available Tools

Search: ls_search_nodes, ls_search_embeddings Read: ls_get_nodes, ls_get_context, ls_query_edges Write: ls_add_node, ls_update_node, ls_create_edge, ls_update_edge Explore: ls_list_dimensions, ls_create_dimension, ls_update_dimension, ls_delete_dimension, ls_sqlite_query Skills: ls_list_skills, ls_read_skill, ls_write_skill, ls_delete_skill

Tips

• Start with ls_read_skill("agent") to understand the operating policy
• Use ls_get_context to get live stats and top nodes
• Use ls_search_embeddings for deep searches through transcripts and articles
• Use ls_sqlite_query for advanced filtering and aggregation
• Always search before creating nodes to avoid duplicates