---

name: server-configuration-and-memory-engines description: "Use the MemMachine server, configuration, REST API, MCP entry points, memory engines, storage/provider resources, Docker/Helm deployment shape, and backend troubleshooting." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Server Configuration And Memory Engines

Use this sub-skill when a task names memmachine-server, MemMachine REST API, MCP stdio/HTTP, cfg.yml, Docker Compose, Helm, storage resources, provider resources, episodic memory, semantic memory, short-term memory, retrieval-agent mode, or server startup/troubleshooting.

Route Within This Sub-skill

• Read server-configuration.md for configuration sections, startup modes, environment variables, and safe validation.
• Read storage-and-providers.md for database/vector/graph stores, embedders, language models, rerankers, optional extras, and credential boundaries.
• Read rest-and-mcp.md for REST endpoint families, health/metrics, MCP tools, and context parameters.
• Read memory-engines.md for event vs declarative long-term memory, short-term memory, semantic/profile memory, and retrieval-agent behavior.
• Read troubleshooting.md for server startup, config, Docker, storage, provider, MCP, and optional dependency failures.
• Run server_config_doctor.py to validate a MemMachine YAML config shape without starting services.
• Run mcp_entrypoint_smoke.py to inspect MCP entry-point availability and context parameters without launching a listener.

Safe Startup Checks

A local server task usually starts with read-only checks:

memmachine-server --help
memmachine-mcp-http --help
python scripts/server_config_doctor.py --config cfg.yml
python scripts/mcp_entrypoint_smoke.py --check-imports

Only start or stop services after the user confirms the desired deployment and side effects. Docker Compose and Helm workflows can create volumes, containers, network listeners, and persistent data.

Configuration Shape

A typical server config has these top-level sections:

logging: {}
episode_store: {}
episodic_memory: {}
retrieval_agent: {}
semantic_memory: {}
session_manager: {}
resources:
  databases: {}
  embedders: {}
  language_models: {}
  rerankers: {}

Important backend decision:

• episodic_memory.long_term_memory.backend: declarative uses a vector_graph_store resource such as Neo4j/Nebula.
• backend: event uses a vector_store plus a segment_store resource.

Server Commands And Entry Points

• memmachine-server --help shows server flags including --stdio, --version, and --with-config-api.
• memmachine-mcp-stdio runs MCP in stdio mode for local MCP clients.
• memmachine-mcp-http --host HOST --port PORT runs MCP over HTTP.
• memmachine-configure is interactive; do not run it in automation unless the user explicitly wants an interactive installer/configurer.

Cross-links

• Use python-sdk-and-cli for application-side Python client calls, mem-cli, and LangGraph helper usage.
• Use integrations-and-migration for framework/platform recipes and conversation import tools.
• Use typescript-rest-client for Node/TS client calls against a MemMachine server.
• Use the root REST API and data models reference when translating endpoint payloads across clients.

Verification Boundaries

• Import and help checks prove packages and entry points are available; they do not prove databases, providers, or MCP clients are configured.
• Docker, Neo4j, Postgres, Qdrant, Milvus, Nebula, OpenAI, AWS Bedrock, Cohere, and Ollama/OpenAI-compatible endpoints are optional external dependencies.
• The optional spaCy multi-hop decomposer may be absent in base installs. Do not classify its warning as a fatal server failure unless the user specifically needs spaCy-based decomposition.