Pinecone MCP Server

Inserts or updates vectors in Pinecone indexes with associated metadata through MCP tool protocol. Implements batch upsert operations that accept vector embeddings, IDs, and structured metadata (key-value pairs), routing them to the Pinecone API with automatic namespace and index targeting. Supports sparse-dense hybrid vectors and metadata filtering for later retrieval.

Queries vectors in Pinecone by semantic similarity using a query vector, returning top-K nearest neighbors with optional metadata filtering. Implements server-side filtering through Pinecone's metadata filter DSL, allowing complex boolean queries (e.g., 'source == "docs" AND date > 2024-01-01') to narrow results before ranking. Supports both dense and sparse-dense hybrid search modes.

Creates, deletes, and describes Pinecone indexes through MCP tools. Handles index configuration (dimension, metric type, pod type, replicas) and provides introspection into index stats (vector count, dimension, metric). Implements index creation with configurable parameters for different workload types (standard, performance, cost-optimized).

Partitions vectors within a single Pinecone index into isolated namespaces, enabling multi-tenant or multi-project data separation without creating separate indexes. Implements namespace targeting in upsert and query operations, allowing vectors with the same ID to coexist in different namespaces. Supports namespace-scoped operations for data isolation and cost optimization.

Deletes vectors from a Pinecone index by ID or metadata filter, supporting both targeted removal and bulk deletion operations. Implements server-side filtering to delete vectors matching metadata criteria (e.g., 'source == "old_docs"'), or direct ID-based deletion for precise removal. Supports namespace-scoped deletion to remove data for a specific tenant or project.

Inspects and describes the metadata schema of vectors in a Pinecone index, returning information about metadata field types, cardinality, and usage patterns. Provides visibility into what metadata fields are present, their data types (string, number, boolean), and how many vectors use each field. Enables schema discovery without manual documentation.

Validates that query and upsert vectors match the index's configured dimension before sending to Pinecone, catching dimension mismatches early in the MCP layer. Implements client-side validation that compares vector length against index metadata, returning clear error messages for dimension mismatches. Prevents wasted API calls and cryptic Pinecone errors.

Automatically generates MCP-compliant tool schemas for all Pinecone operations (upsert, query, delete, index management), enabling seamless integration with MCP clients like Claude. Implements schema generation that includes input/output types, descriptions, and required parameters, following MCP specification for tool calling. Allows LLM agents to discover and use Pinecone operations without manual schema definition.

Implements automatic retry logic for transient Pinecone API failures (rate limits, temporary outages) with exponential backoff, and provides detailed error messages for permanent failures (invalid API key, dimension mismatch). Handles MCP-specific error responses that include error codes, messages, and recovery suggestions. Reduces noise from transient failures while surfacing actionable errors.

Manages Pinecone API authentication through environment variables or configuration files, supporting secure credential storage without embedding keys in code. Implements credential validation at startup and automatic re-authentication if tokens expire. Supports both API key and OAuth-based authentication (if Pinecone supports it), with fallback mechanisms for missing credentials.