Couchbase

Converts natural language questions into Couchbase N1QL (SQL-like query language) statements through LLM-powered semantic understanding. The MCP server acts as an intermediary that parses user intent, constructs appropriate N1QL syntax with proper bucket/scope/collection references, and executes against connected Couchbase clusters. This enables non-SQL developers to query document databases using conversational language without learning N1QL syntax.

Automatically discovers and catalogs Couchbase cluster structure including buckets, scopes, collections, indexes, and document schemas through direct cluster API calls. The MCP server queries system catalogs and samples documents to build a schema model that can be injected into LLM context, enabling accurate natural language query generation and reducing hallucination about field names and data structures.

Executes pre-written or generated N1QL queries directly against Couchbase clusters and streams results back through the MCP protocol. The server maintains connection pooling to the cluster, handles query timeouts and retries, and formats results as JSON for consumption by LLM agents or client applications. Supports parameterized queries to prevent injection attacks and enable safe dynamic query construction.

Provides atomic read, insert, update, and delete operations on individual Couchbase documents through MCP tool bindings. Supports optimistic concurrency control via CAS (Compare-And-Swap) tokens to prevent lost updates in concurrent scenarios, and allows specification of consistency levels (eventual, strong) for read operations. Operations are transactional at the document level and can be chained in agent workflows.

Executes Couchbase Full-Text Search queries through MCP tools, enabling semantic and keyword-based document retrieval across large collections. The server translates search criteria into FTS query syntax, handles faceting and result ranking, and returns ranked results with relevance scores. Supports complex queries including boolean operators, phrase search, and field-specific search within indexed documents.

Executes multiple document operations (inserts, updates, deletes) in a single batch request with per-document error handling and partial success reporting. The server optimizes batch operations for throughput using connection pooling and pipelining, and returns detailed results indicating which operations succeeded and which failed with specific error reasons. Useful for bulk data loading or multi-document mutations from agent workflows.

Caches N1QL query results in memory with configurable TTL and provides cursor-based pagination for large result sets. The server maintains a result cache indexed by query hash, enabling repeated queries to return cached results without re-executing against the cluster. Pagination uses cursor tokens to maintain position across multiple requests, avoiding offset-based inefficiency for large datasets.

Monitors Couchbase cluster health by querying node status, service availability, bucket statistics, and query performance metrics. The MCP server exposes cluster diagnostics as tools that agents can invoke to validate cluster state before executing queries, detect performance issues, or report health status. Includes metrics like memory usage, replication lag, and query queue depth.