MCP tools¶

memini-ai exposes 52 MCP tools grouped by capability. All tools are available over stdio (memini-ai --stdio) once your MCP client connects.

Foundation memory¶

Tool	Description
`add_memory`	Store a new memory with optional source-type and metadata.
`query_memories`	Semantic search with `tiered` / `vector_only` / `text_only` / `parallel` strategies.
`get_memory`	Fetch a single memory by ID.
`delete_memory`	Remove a memory entry by ID.

Project indexing¶

Tool	Description
`index_project`	Trigger recursive indexing of a project directory.
`search_project`	Semantic search across indexed project files.
`get_file_contents`	Reconstruct a file from its indexed chunks.
`get_indexing_status`	Check progress of a background indexing job.

Trust engine¶

Tool	Description
`get_trust_score`	Retrieve the trust score and level for a memory.
`adjust_trust`	Apply a feedback signal (`agent_used`, `agent_ignored`, `user_confirmed`, `user_corrected`).
`adjust_decay_rate`	Change the decay rate for a specific memory (sticky = low rate).
`get_decay_status`	View the decay engine status and fading memories.
`list_fading_memories`	List memories approaching the archive threshold.
`list_archived`	List memories already archived (below archive threshold).

Tiered loading¶

Tool	Description
`get_tier0_summary`	High-trust L0 project summary (~100 tokens).
`get_tier1_summary`	L1 key-decisions summary (~2K tokens, trust >= 0.8).

Knowledge graph¶

Tool	Description
`query_kg`	Execute a formal knowledge-graph query.
`extract_entities`	Extract named entities from a memory entry.
`search_entities`	Find entities by name.
`get_entity_graph`	Get all connections to/from an entity.
`get_inference_chain`	Find inference paths between two entities.
`get_graph_visualization`	Get a self-contained HTML page with a D3.js force-directed graph.

Memory relationships¶

Tool	Description
`find_related_memories`	Find memories linked to a reference memory (`SUPERSEDES`, `PARTIAL_UPDATE`, etc.).
`get_relationship_summary`	Get a summary of all relationships for a memory.
`create_relationship`	Create a typed relationship between two memories.
`trigger_consolidation`	Manually merge similar memory pairs.

Dialectic¶

Tool	Description
`find_contradictions`	Detect memory pairs that contradict each other.
`resolve_contradiction`	Generate an LLM-synthesized dialectic resolution for two memories.
`challenge_memory`	Submit a counter-argument challenge to a memory.
`get_dialectic_history`	View the argument history for a memory.

Thought chains¶

Tool	Description
`start_thought_chain`	Create a new reasoning chain (optionally parented to another chain).
`add_thought`	Add a thought to a chain (auto-creates the chain if no `chain_id` is given).
`branch_thought`	Start a new branch from an existing thought.
`revise_thought`	Create a revision of an existing thought.
`get_thought_chain`	Retrieve a full chain with all thoughts organized by branch.
`get_related_chains`	Search for thought chains with reasoning similar to a query.
`pause_thought_chain`	Pause a chain.
`resume_thought_chain`	Resume a paused chain (returns the last thought).
`abandon_thought_chain`	Mark a reasoning path as incorrect.

Multi-peer¶

Tool	Description
`list_peers`	List all known peers.
`add_peer`	Register a new peer with a role (`owner`, `collaborator`, `readonly`, `guest`).
`switch_peer_context`	Switch the active peer context for memory operations.
`share_memory`	Share a memory with another peer (`shared` or `inherited` permission).
`get_peer_memories`	Query another peer's accessible memories.
`get_shared_memories`	Get all memories shared with the current peer context.

User modeling¶

Tool	Description
`get_user_profile`	Retrieve the learned user profile and style.
`update_user_profile`	Update the profile from the current conversation.

Extraction¶

Tool	Description
`trigger_extraction`	Manually trigger memory extraction from a conversation.
`preconpress_extraction`	Capture context and extract memories before compaction.

Dual-model RRF¶

Tool	Description
`elevate_memory_to_1024`	Promote a 384-dim memory into 1024-dim space (auto-mode only).

System and maintenance¶

Tool	Description
`get_status`	Health check for all server components (reports `modelName`, `modelDimension`, `embeddingDimMismatch`).
`healthcheck`	End-to-end write + read round-trip probe with latency metrics.
`log_audit_event`	Manually log an audit event.
`get_audit_log`	Query the audit log with filters.
`get_security_summary`	Aggregated security metrics for the last N hours.

Controlling which tools are exposed¶

memini-ai registers 52 MCP tools by default. You can reduce the exposed surface so your MCP client sends fewer tool schemas (saving tokens per request). There are two mechanisms, and the more specific one wins:

1. Per-tool allow-list — `MEMINI_ENABLED_TOOLS`¶

When set, only the named tools are registered. This overrides MEMINI_TOOL_GROUPS.

export MEMINI_ENABLED_TOOLS="query_memories,add_memory"

Or in opencode.json:

{
  "mcp": {
    "servers": {
      "memini-ai-dev": {
        "type": "local",
        "enabled": true,
        "environment": {
          "MEMINI_ENABLED_TOOLS": "query_memories,add_memory"
        }
      }
    }
  }
}

Rules:

Names are the exact function names used at registration time.
Whitespace around commas is trimmed.
Unknown names are ignored with a warning log.
If unset/empty, the server falls back to MEMINI_TOOL_GROUPS.

2. Group-based gating — `MEMINI_TOOL_GROUPS`¶

Coarse groups exposed when MEMINI_ENABLED_TOOLS is unset. Default: core,trust,kanban,session.

Group	Tools
`core`	`add_memory`, `query_memories`, `get_memory`, `delete_memory`
`trust`	trust-engine tools
`kanban`	kanban board tools
`session`	`get_status`, `healthcheck`, `trigger_compaction`, `trigger_consolidation`
`chains`	thought-chain tools
`kg`	knowledge-graph tools
`dialectic`	contradiction-resolution tools
`peers`	multi-peer sharing tools
`memory_ops`	relationship tools
`audit`	audit-log tools
`ops`	project-indexing tools

Full details and examples are in Configuration.