MCP Tools Reference

Complete reference for all 11 tools exposed by the Kronroe MCP server. Each tool is callable via the Model Context Protocol stdio transport.

remember

Ingest free-text and store extracted facts in memory.

Parameter	Type	Required	Description
text	string	Yes	The text to parse and store as facts. Max 32 KiB.
episode_id	string	No	Group facts under a conversation or episode. Defaults to "default". Max 512 bytes.
idempotency_key	string	No	Deduplicate repeated ingestion of the same text. Cannot be combined with query_embedding. Max 512 bytes.
query_embedding	array of numbers	No	Pre-computed embedding vector for the text. Requires the hybrid feature. Cannot be combined with idempotency_key.

Example:

{
  "text": "Alice joined Acme Corp in January 2025 as a senior engineer.",
  "episode_id": "onboarding-chat"
}

recall

Recall facts by natural-language query using full-text search.

Parameter	Type	Required	Description
query	string	Yes	Natural-language search query. Max 8 KiB.
limit	integer	No	Maximum number of results to return. Range: 1–200. Default: 10.
include_scores	boolean	No	When true, returns per-channel scoring metadata alongside each fact. Default: false.
min_confidence	number	No	Minimum confidence threshold for returned facts. Range: 0.0–1.0.
confidence_filter_mode	string	No	Which confidence signal to filter on: "base" (raw fact confidence) or "effective" (uncertainty-aware; requires uncertainty feature). Requires min_confidence to be set.
max_scored_rows	integer	No	Limit the number of rows that receive full scoring computation. Minimum: 1.
query_embedding	array of numbers	No	Pre-computed embedding vector for hybrid retrieval. Requires the hybrid feature.
use_hybrid	boolean	No	Enable hybrid (text + vector) retrieval. Requires query_embedding and the hybrid feature.
temporal_intent	string	No	Temporal intent hint for the reranker: "timeless", "current_state", "historical_point", or "historical_interval". Requires the hybrid feature.
temporal_operator	string	No	Temporal operator hint: "current", "as_of", "during", "before", "by", "after", or "unknown". Requires the hybrid feature.

Example:

{
  "query": "where does Alice work",
  "limit": 5,
  "min_confidence": 0.8
}

recall_scored

Recall facts with per-channel scoring metadata included for every result. Identical parameters to recall, except there is no include_scores parameter (scores are always included).

Parameter	Type	Required	Description
query	string	Yes	Natural-language search query. Max 8 KiB.
limit	integer	No	Maximum number of results. Range: 1–200. Default: 10.
min_confidence	number	No	Minimum confidence threshold. Range: 0.0–1.0.
confidence_filter_mode	string	No	"base" or "effective". Requires min_confidence.
max_scored_rows	integer	No	Limit rows receiving full scoring. Minimum: 1.
query_embedding	array of numbers	No	Pre-computed embedding vector. Requires the hybrid feature.
use_hybrid	boolean	No	Enable hybrid retrieval. Requires query_embedding and the hybrid feature.
temporal_intent	string	No	Temporal intent hint. Requires the hybrid feature.
temporal_operator	string	No	Temporal operator hint. Requires the hybrid feature.

Each result includes a score breakdown with fields like rrf_score, text_contrib, vector_contrib, confidence, and effective_confidence.

Example:

{
  "query": "Alice's job title",
  "limit": 3
}

assemble_context

Build LLM-ready context text from the top-ranked facts matching a query, constrained by a token budget.

Parameter	Type	Required	Description
query	string	Yes	Natural-language query to find relevant facts. Max 8 KiB.
max_tokens	integer	Yes	Maximum token budget for the assembled context. Minimum: 1.
query_embedding	array of numbers	No	Pre-computed embedding vector for hybrid retrieval.

Example:

{
  "query": "everything about Alice's career",
  "max_tokens": 500
}

facts_about

Return all current facts about a specific entity.

Parameter	Type	Required	Description
entity	string	Yes	The entity name to look up (e.g., "alice").

Example:

{
  "entity": "alice"
}

assert_fact

Assert a structured fact with explicit subject, predicate, and object.

Parameter	Type	Required	Description
subject	string	Yes	The entity the fact is about (e.g., "alice").
predicate	string	Yes	The relationship or property name (e.g., "works_at").
object	any	Yes	The value: a string, number, or boolean.
valid_from	string	No	RFC 3339 timestamp for when this fact became true. Defaults to the current time.
confidence	number	No	Confidence score. Range: 0.0–1.0. Default: 1.0. Cannot be combined with idempotency_key.
source	string	No	Provenance marker (e.g., "user_statement"). Cannot be combined with idempotency_key.
idempotency_key	string	No	Deduplicate repeated assertions. Cannot be combined with confidence or source.

Example:

{
  "subject": "alice",
  "predicate": "works_at",
  "object": "Acme Corp",
  "confidence": 0.95,
  "source": "user_statement",
  "valid_from": "2025-01-15T00:00:00Z"
}

correct_fact

Correct a fact’s value by its ID. The old fact is preserved in history with its validity window closed, and a new fact is created with the corrected value.

Parameter	Type	Required	Description
fact_id	string	Yes	The Kronroe Fact ID (kf_...) of the fact to correct.
new_value	any	Yes	The corrected value (string, number, or boolean).

Example:

{
  "fact_id": "kf_01HX...",
  "new_value": "Globex Corp"
}

invalidate_fact

Invalidate a fact by its ID, ending its validity window. The fact is not deleted – it remains in history with valid_to set to the current time.

Parameter	Type	Required	Description
fact_id	string	Yes	The Kronroe Fact ID (kf_...) of the fact to invalidate.

Example:

{
  "fact_id": "kf_01HX..."
}

what_changed

Return a change report for an entity since a given timestamp. Shows new facts, invalidated facts, and corrections.

Parameter	Type	Required	Description
entity	string	Yes	The entity to check for changes.
since	string	Yes	RFC 3339 timestamp. Only changes after this time are included.
predicate	string	No	Filter changes to a specific predicate (e.g., "works_at").

Example:

{
  "entity": "alice",
  "since": "2025-01-01T00:00:00Z",
  "predicate": "works_at"
}

memory_health

Return an operational health report for an entity’s stored facts. Identifies low-confidence facts, stale high-impact facts, and contradictions.

Parameter	Type	Required	Description
entity	string	Yes	The entity to assess.
predicate	string	No	Scope the report to a specific predicate.
low_confidence_threshold	number	No	Facts with confidence below this value are flagged. Range: 0.0–1.0. Default: 0.7.
stale_after_days	integer	No	High-impact facts older than this many days are flagged as stale. Minimum: 0. Default: 90.

Example:

{
  "entity": "alice",
  "low_confidence_threshold": 0.5,
  "stale_after_days": 180
}

recall_for_task

Return decision-ready memory context scoped to a specific task. Provides key facts, watchouts (low-confidence or stale facts relevant to the task), and recommended next checks.

Parameter	Type	Required	Description
task	string	Yes	Description of the task or decision to prepare for. Max 8 KiB.
subject	string	No	Scope recall to a specific entity.
now	string	No	RFC 3339 timestamp to use as “now” for staleness calculations. Defaults to the current time.
horizon_days	integer	No	How far back to look for relevant facts, in days. Minimum: 1. Default: 30.
limit	integer	No	Maximum number of key facts to return. Range: 1–200. Default: 8.
query_embedding	array of numbers	No	Pre-computed embedding vector for hybrid retrieval. Requires the hybrid feature.
use_hybrid	boolean	No	Enable hybrid retrieval. Requires query_embedding and the hybrid feature.

Example:

{
  "task": "Write a performance review for Alice",
  "subject": "alice",
  "horizon_days": 365,
  "limit": 10
}