MCP Tools

RushDB MCP exposes 43 tools on RushDB Cloud; in direct mode against a self-hosted engine, the 6 Cloud-only tools are not listed. Names below use the exact lower camel case identifiers published in the MCP manifest.

The connected MCP client receives the full JSON schema for each tool. This page is the quick reference for choosing the correct tool and understanding its main inputs.

Safety Levels

Level	Meaning
Read	Retrieves or computes data without changing the data
Write	Creates or updates project data
Destructive	Deletes or replaces data; preview and confirm first

An asterisk marks a required input.

Discovery and Properties

Tool	Level	Main inputs	Purpose
getSchemaMarkdown	Read	labels?, force?	Return the complete graph schema as compact Markdown. Call once at session start.
getSchema	Read	labels?, force?	Return structured schema JSON, including property IDs and vector indexes.
findLabels	Read	where?, limit?, skip?, orderBy?	List or filter record labels and counts.
findProperties	Read	where?, limit?, skip?, orderBy?	Discover property names, types, IDs, and record counts.
findPropertyById	Read	propertyId*	Fetch metadata for one property.
propertyValues	Read	propertyId*, query?, orderBy?, limit?, skip?	Return numeric or date ranges, or distinct string and boolean values.
deleteProperty	Destructive	propertyId*	Permanently remove a property and its values from all records.
getSearchQuerySpec	Read	none	Return the complete SearchQuery syntax reference.
getQueryBuilderPrompt	Read	none	Return the built-in discovery-first query builder prompt.

Records

Tool	Level	Main inputs	Purpose
findRecords	Read	labels?, where?, limit?, skip?, orderBy?, select?, groupBy?	Search records and run grouped metrics queries.
findOneRecord	Read	labels?, where?	Return the first matching record for entity resolution probes.
findUniqRecord	Read	labels?, where?	Return exactly one matching record; fail when zero or multiple records match.
getRecord	Read	recordId*	Fetch one record by ID.
getRecordsByIds	Read	recordIds*	Fetch multiple records by ID.
createRecord	Write	label*, data*, transactionId?, options?	Insert or upsert one record.
updateRecord	Write	recordId*, label*, data*, transactionId?	Partially update one record while preserving unspecified fields.
setRecord	Destructive	recordId*, label*, data*, transactionId?	Replace all fields on one record.
deleteRecord	Destructive	recordId*, transactionId?	Delete one record by ID after previewing it with findRecords.
deleteRecordById	Destructive	recordId*, transactionId?	Delete one record by ID after previewing it with getRecord.
bulkCreateRecords	Write	label*, data*, transactionId?, options?	Insert or upsert multiple records with the same label.
bulkDeleteRecords	Destructive	labels?, where*, transactionId?	Delete all records matching a query. Preview with findRecords first.
exportRecords	Read	labels?, where?, limit?, orderBy?	Export matching records as CSV.

Relationships

Tool	Level	Main inputs	Purpose
findRelationships	Read	where?, source?, target?, limit?, skip?, orderBy?	Inspect edges. where filters edge type/properties; source/target filter endpoint records.
attachRelation	Write	sourceId*, targetId?, targetIds?, relationType?, direction?, properties?, transactionId?	Create relationships between existing records.
detachRelation	Destructive	sourceId*, targetId?, targetIds?, relationType?, direction?	Remove relationships between records: every type unless relationType is given, both ways unless direction is.

Vector, Keyword and Hybrid Search

Tool	Level	Main inputs	Purpose
findEmbeddingIndexes	Read	none	List embedding indexes with their names and status.
createEmbeddingIndex	Write	name?, label*, propertyName*, sourceType?, similarityFunction?, dimensions?	Create an embedding index. On a self-hosted engine, create an external one (sourceType: "external", dimensions).
getEmbeddingIndexStats	Read	indexId*	Check how many records an index covers and has vectors for.
upsertEmbeddingVectors	Write	indexId*, items*	Write pre-computed vectors to an external index.
deleteEmbeddingIndex	Destructive	indexId*	Delete an embedding index and its stored vectors.
vectorSearch	Read	index?, propertyName?, query?, queryVector?, labels?, sourceType?, similarityFunction?, dimensions?, where?, mode?, ef?, scoreThreshold?, weight?, prefetchLimit?, fusion?, rrfK?, text?, limit?, skip?	Rank records by vector similarity after exact filtering; add text for a hybrid search. On a self-hosted engine, pass queryVector.
semanticSearch	Read	propertyName*, query?, queryVector?, labels*, sourceType?, similarityFunction?, dimensions?, where?, mode?, ef?, scoreThreshold?, limit?, skip?	Deprecated alias for vectorSearch.
listTextIndexes	Read	none	List text indexes with their status and pending changes.
createTextIndex	Write	label*, property*, name?, consistency?	Declare a BM25 text index over one property, or * for every string.
deleteTextIndex	Destructive	name*	Drop a text index by name.
textSearch	Read	query*, index?, property?, labels?, where?, allTerms?, phrase?, limit?, skip?	Keyword (BM25) search over a text index, after exact filtering.

RushDB Cloud: Smart Search and Relationship Patterns

These tools need RushDB Cloud. In direct mode against a self-hosted engine they are not listed, and a call is refused.

Tool	Level	Main inputs	Purpose
smartSearch	Read	prompt*, currentQuery?	Turn a natural-language request into a SearchQuery and run it.
listRelationshipPatterns	Read	none	List inferred patterns, schema relationship summaries, and analysis status.
analyzeRelationshipPatterns	Write	none	Queue schema analysis with RushDB Cloud's language model.
approveRelationshipPattern	Write	id*	Approve and apply a suggested pattern.
ignoreRelationshipPattern	Write	id*	Ignore a suggested pattern without applying it.
deleteRelationshipPattern	Destructive	id*, deleteExisting?	Delete a stored pattern and optionally its materialized relationships.

Utilities

Tool	Level	Main inputs	Purpose
helpAddToClient	Read	none	Return instructions for adding RushDB MCP to a client.

Query Rules

• Never guess labels or property names. Start with getSchemaMarkdown.
• Call getSearchQuerySpec before complex findRecords queries.
• Do not pass limit with select; it restricts the record scan and produces incorrect metrics.
• For a simple count, read total from findRecords.
• Preview and explicitly confirm destructive operations.