Reference

MCP Tools

All tools exposed by the MemoryLayer MCP server — collapsed by default, filterable by category.

All tool categories write to the same local SQLite store.

Tip

Eight tools, one unified surface: remember (write), recall (read), code (code intelligence), consolidate (maintenance + ACT-R compose), verify (hallucination gate), prove (Z3 formal proof), domain (axiom management), audit (batch invariant check).

Write path — store, update, delete, or bulk-import memories. Routes internally to the right operation based on params. `upsert` is the default and the one to reach for: it dedupes near-duplicates, extracts CALLS/IS-A/IMPORTS triples (so relational queries work later), and returns a contradiction warning if a conflicting fact already exists. `chunk` and `batch` require Pro+.

Actions

upsert (default)Store or update in place. Near-duplicates update; new content creates. Phrase content relationally ('AuthService calls verify_token') to enable structural queries.

updateEdit an existing memory by id — re-embeds if content changes.

deletePermanently remove one memory by id.

chunkPro+ onlyStore a long document as overlapping searchable chunks. Ideal for RFCs, ADRs, long transcripts.

batchPro+ onlyStore up to 100 memories in one call.

Parameters

contentstringoptionalText to remember. Required for upsert/chunk; one item in batch.

namespacestringoptionalLogical group. Default: 'default'

tagsstring[]optionalLabels for later filtering

priorityinteger 0–10optionalRetrieval priority (default 5). Use 8–10 for hard constraints.

metadataobjectoptionalArbitrary key-value pairs. Set metadata.skill_name to enable skill_load by name.

idstringoptionalMemory ID — required for update and delete actions.

action'upsert'|'update'|'delete'|'chunk'|'batch'optionalDefaults to 'upsert' when content is provided without an id.

similarity_thresholdnumberoptionalDedup strictness 0–1 (default ~0.85). Lower = more aggressive dedup.

chunk_sizenumberoptionalWords per chunk for action:'chunk'

itemsobject[]optionalArray of memory objects for action:'batch'. Max 100.

Returns

upsert → { id, namespace, action: "created"|"updated", triples_extracted[]?, contradiction? } | chunk → { parent_id, chunk_ids[], chunk_count } | batch → { stored, results[] } | update/delete → { id, status }

Example

// Upsert (default) — dedupes, extracts triples
remember({
  content: "AuthService calls verify_token to validate JWTs.",
  namespace: "myproject",
  tags: ["auth"]
})
// → { action: "created", triples_extracted: [{ s: "AuthService", p: "CALLS", o: "verify_token" }] }

// Update
remember({ action: "update", id: "abc123", content: "Updated." })

// Delete
remember({ action: "delete", id: "abc123" })

// Chunk a long doc (Pro+)
remember({ action: "chunk", content: "<RFC>", namespace: "docs" })

// Batch (Pro+)
remember({ action: "batch", items: [{ content: "Fact one" }, { content: "Fact two" }] })

Rate limits

Plan	Calls / month	Price
Free	1,000	$0
Pro	5,000	$19/mo
Pro+	10,000	$49/mo
Enterprise	custom	contact

Limits reset on the 1st of each month (UTC). See the pricing page for per-tool quotas.

// Upsert (default) — dedupes, extracts triples remember({ content: "AuthService calls verify_token to validate JWTs.", namespace: "myproject", tags: ["auth"] }) // → { action: "created", triples_extracted: [{ s: "AuthService", p: "CALLS", o: "verify_token" }] } // Update remember({ action: "update", id: "abc123", content: "Updated." }) // Delete remember({ action: "delete", id: "abc123" }) // Chunk a long doc (Pro+) remember({ action: "chunk", content: "<RFC>", namespace: "docs" }) // Batch (Pro+) remember({ action: "batch", items: [{ content: "Fact one" }, { content: "Fact two" }] })

Plan

Calls / month

Price

Free

1,000

Pro

5,000

$19/mo

Pro+

10,000

$49/mo

Enterprise

custom

contact