MCP Tools

Complete reference for 3ngram’s 18 MCP tools. Parameters, read/write classification, and usage examples for the AI memory server.

3ngram’s MCP server provides 18 tools for AI assistants to interact with the knowledge base.

Tool catalog

Tool	Purpose	R/W
remember	Persist a structured memory with type, scope, project, due date	Write
recall	Retrieve memories by semantic similarity	Read
get_memories	Fetch full memory content by IDs	Read
resolve	Mark a memory as resolved with optional note	Write
commitments	List open commitments	Read
blockers	List open blockers	Read
overdue	List commitments past their deadline	Read
stale_commitments	List commitments with no activity for N+ days	Read
search	Hybrid semantic + BM25 search across content chunks	Read
reindex	Re-index markdown content into content chunks	Write
reclassify_memory	Change the type of a memory	Write
suggest_reclassifications	Find context memories that may be misclassified	Read
triage_stale	Surface stale memories with resolution prompts	Read
archive_memories	Archive resolved or stale memories by ID	Write
handoff	Export session context as structured JSON	Read
configure_scope	Assign a project to a scope (work/personal)	Write
reapply_scope_mappings	Reapply scope mappings to existing memories	Write
status	Index statistics and memory breakdown	Read

Key parameters

remember

Parameter	Type	Required	Description
topic	string	Yes	Short identifier for the memory
content	string	Yes	Full memory content
memory_type	string	No	decision, commitment, blocker, preference, pattern, context
project	string	No	Project tag
scope	string	No	work or personal
due_date	string	No	ISO date for commitments
status	string	No	open, waiting, scheduled

recall

Parameter	Type	Required	Description
topic	string	Yes	Search query
brief	boolean	No	Compact one-line-per-memory format
limit	integer	No	Max results (default 10)
memory_type	string	No	Filter by type
project	string	No	Filter by project
scope	string	No	Filter by scope
source	string	No	Filter by source (manual, hook, auto)

resolve

Parameter	Type	Required	Description
memory_id	integer	Yes	Memory to resolve
resolution_note	string	No	What happened

Output format

All tools return human-readable formatted strings, not raw data. Example commitments output:

Open Commitments (3):
  [122] Implement memory consolidation UI (5 days) [project: engram]
  [125] Write integration tests for OAuth flow (3 days) [project: engram]
  [131] Review Cursor MCP configuration (1 day) [project: climbr]

Progressive disclosure

Use recall(brief=True) to scan, then get_memories(ids=[...]) to read details. This keeps token usage low when browsing large memory sets.

Auto-consolidation

When remember finds an existing memory with 85%+ similarity, it merges instead of creating a duplicate. Consolidation uses advisory locks to prevent race conditions.