initialize-knowledge-graph

Initialize Knowledge Graph

Automate the discovery of domain terminology and codebase structure, translating them into persistent nodes (Entities), edges (Relations), and facts (Observations) inside the Memory MCP server.

This ensures the agent starts with high-signal context and complete terminology alignment without requiring manual explanation.

Process

1. Discovery & Research

Scan the codebase to gather high-level domain information and structural entry points:

• Domain Docs: Look for CONTEXT.md, CONTEXT-MAP.md, and any files in docs/adr/.
• Glossary & Jargon: Extract canonical terms, abbreviations, and descriptions defined in these files.
• Codebase Structure: Scan major modules, entry points, or directory layout using get_repo_outline or get_file_tree.

2. Formulate Entity List

Before writing to the graph, list all identified entities and their relationships. Group them into:

• Concepts: Core business domains or components (e.g. Character card, Persona card).
• Actors/Roles: Active agents or user roles in the system (e.g. User, Bot).
• Structures: Physical or logical data layouts (e.g. Lorebook, Database).

3. Populate the Graph (Memory Ingestion)

Proactively call the memory MCP server mutation tools to build the knowledge graph structure:

• Create Entities: Call create_entities to add all mapped concepts and actors. Include a brief, precise description in each node's observations.
• Set Relations: Call create_relations to establish structural and semantic connections.
• Configure Aliases (CRITICAL): For any acronyms, shortcuts, or alternate names, create a separate Entity and link it to the canonical Entity using an is_alias_of or synonym_of relation.
  • Example: Create Auth as a Concept, and establish Auth $\rightarrow$ is_alias_of $\rightarrow$ Authentication.

4. Verify & Confirm

Validate that the knowledge graph is correctly populated:

• Call read_graph or perform a search_nodes query on major terms to verify the nodes are correctly linked and queryable.
• Present a concise, professional summary of the initialized entities and relationships to the user.