MCP Tools Overview
BaseLayer exposes eleven tools via MCP. Any connected AI can call these to interact with your memory.Querying
| Tool | Purpose | When to use |
|---|---|---|
| ask_question | Synthesize answers with citations | Factual questions about you, your projects, preferences, and history |
Searching & Exploring
| Tool | Purpose | When to use |
|---|---|---|
| memory_search | Semantic search across entities and facts | Browsing, exploring, or discovering entities in your memory |
| get_entity | Fetch a full entity dossier | Deep dive into a specific person, project, plan, or concept |
| get_entity_relations | Browse graph edges and connections | Understanding how entities are linked, filtering connections |
| get_entity_provenance | Track fact origins | Verifying where knowledge came from, auditing sources |
| list_entities | Browse all entities of a type | Discovering what’s in your memory, finding plans |
| recent_conversations | Load recent dialogue context | Start of a conversation, catching up on recent activity |
| retrieve_evidence | Search raw conversation windows | Finding exact quotes, code snippets, or original wording |
Writing
| Tool | Purpose | When to use |
|---|---|---|
| record_memory | Persist observations and context | Saving decisions, notes, corrections, preferences |
| vault_fact | Assert precise facts in the graph | Creating specific entity relationships and edges |
| update_plan | Persistent markdown scratchpads | Tracking multi-step efforts across sessions |
Common patterns
Start of conversation: The AI callsrecent_conversations to load recent context, then uses ask_question as needed throughout the conversation.
Factual questions: ask_question first for a synthesized answer with citations, then get_entity if more detail is needed on a specific entity.
Research mode: memory_search to discover entities, then get_entity for full dossiers, then get_entity_relations to explore connections.
Storing knowledge: record_memory for general notes and decisions, vault_fact for precise facts linking two entities.
Plans: update_plan to create or update a persistent scratchpad, get_entity(type="plan") to read it back in any future session.
Tool priority order
When an AI needs to find information, it should follow this priority chain:- ask_question — always try this first for any factual query
- memory_search — browse, explore, or find entities by type
- get_entity — read a full dossier before acting on partial search results
- retrieve_evidence — only when exact quotes or raw context are needed
How AIs use these tools
You don’t need to explicitly ask your AI to call these tools. Connected AIs use them naturally based on context:- If you ask about past work, the AI calls
ask_question - If you share a decision, the AI calls
record_memory - If you ask “what do you know about X,” the AI calls
get_entity - If you start a multi-session effort, the AI creates a plan with
update_plan
“Search my memory for anything about the Stripe integration.”
“Remember that we decided to use Redis for caching.”
“Create a plan to track the auth refactor.”