---

name: mcp-server description: "Guides Graphiti MCP tools, transports, config, and smoke checks." metadata: disco-role: operating disable-model-invocation: true license: Apache 2.0

mcp-server

Use this sub-skill for the Graphiti MCP server in mcp_server/: tool names, transport selection, config files, custom entity and edge types, Docker images, and client-level smoke checks.

Read first

• references/mcp-tools.md for the verified tool catalog and payload shapes.
• references/configuration.md for YAML settings, CLI overrides, transports, and backend/provider selection.
• references/troubleshooting.md when the server fails to start, tool calls return errors, or a client is pointed at the wrong transport or backend.
• scripts/mcp_smoke.py for a bundled tool-listing and tiny end-to-end smoke.

What belongs here

Use this sub-skill when the task mentions:

• graphiti_mcp_server, main.py, FastMCP, mcp, stdio, http, or sse
• add_memory, search_nodes, search_memory_facts, get_episodes, get_status, clear_graph, summarize_saga, build_communities, add_triplet, or get_episode_entities
• config.yaml, GraphitiConfig, QueueService, entity_types, edge_types, or edge_type_map
• Docker Compose for the combined FalkorDB container or the Neo4j container path
• Cursor or Claude Desktop MCP integration questions

What does not belong here

Route these elsewhere:

• Direct Graphiti SDK questions -> sub-skills/core-sdk/
• REST route questions and FastAPI deployment -> sub-skills/rest-service/
• Repo maintenance, provenance, or package import details -> root references

MCP workflow

1. Choose a backend and model configuration.
2. Choose a transport: stdio, http, or sse.
3. Start the server.
4. List tools to confirm the expected catalog.
5. Use add_memory or add_triplet to write data.
6. Search with search_nodes or search_memory_facts.
7. Validate with get_status and clean up with clear_graph if needed.

Tool families

• Memory ingest: add_memory, add_triplet
• Search: search_nodes, search_memory_facts
• Retrieval and provenance: get_episodes, get_entity_edge, get_episode_entities
• Maintenance: delete_episode, delete_entity_edge, clear_graph, build_communities, summarize_saga
• Diagnostics: get_status, /health

Validation path

For a local stdio smoke:

python scripts/mcp_smoke.py --transport stdio

For an HTTP smoke against an already running server:

python scripts/mcp_smoke.py --transport http --base-url http://127.0.0.1:8000