Redis MCP Server

Translates conversational AI agent queries into Redis operations through a decorator-based tool registration system built on FastMCP framework. The RedisMCPServer class exposes Redis functionality as MCP tools, allowing agents to express intent in natural language (e.g., 'cache this item') which maps to specific Redis commands (e.g., string.set_string with expiration) without requiring knowledge of Redis syntax or command semantics.

Exposes Redis operations through multiple communication transports configured via MCP_TRANSPORT environment variable: stdio for local development and direct process communication, Server-Sent Events (SSE) for network-based deployments, and containerized Docker deployment with environment-based configuration. The server abstracts transport selection, allowing the same RedisMCPServer implementation to serve different deployment topologies without code changes.

Provides vector search capabilities through a redis_query_engine tool that creates vector indexes, stores embeddings, and performs similarity searches. The engine abstracts Redis Search module operations (HNSW algorithm for approximate nearest neighbor search), enabling agents to index documents with embeddings and retrieve semantically similar results. Agents can express intent like 'index and search this vector' which maps to create_index() + vector_search(), enabling semantic search without understanding vector database internals or similarity metrics.

Implements Redis JSON operations through MCP tools for storing and querying JSON documents using JSONPath expressions. Tools support JSON.SET (store document), JSON.GET (retrieve with path), and JSON.MGET (multi-document retrieval). The JSON module enables structured document storage with path-based access, allowing agents to store complex objects and query nested fields without serialization overhead. Agents can express intent like 'store configuration object' which maps to JSON.SET with automatic JSON validation.

Provides server management tools for database administration through MCP tools exposing INFO (server statistics), FLUSHDB (clear database), DBSIZE (count keys), and configuration operations. Tools enable agents to monitor Redis health, manage database state, and perform administrative tasks. Agents can express intent like 'check Redis health' which maps to INFO command, enabling observability and operational management without direct Redis CLI access.

Manages Redis connections through a RedisConnectionManager singleton pattern that handles both standalone Redis instances and Redis Cluster deployments. The connection manager maintains a pool of connections, handles SSL/TLS encryption, manages authentication via environment variables, and abstracts cluster topology discovery. All MCP tools reference the singleton, ensuring connection reuse and preventing connection exhaustion across concurrent agent requests.

Implements Redis string operations through MCP tools that support basic key-value storage with optional time-to-live (TTL) expiration. The set_string tool accepts key, value, and optional expiration parameters, automatically translating to Redis SET command with EX (seconds) or PX (milliseconds) options. Agents can express intent like 'cache this item' which maps to set_string with appropriate TTL, enabling automatic cache invalidation without explicit deletion logic.

Provides Redis hash operations through MCP tools for storing and manipulating structured data with named fields. Tools like hset, hget, hgetall, and hdel operate on hash keys, allowing agents to store objects as field-value pairs (e.g., storing user profiles with fields like 'name', 'email', 'created_at'). Hashes support TTL at the key level, enabling expiring structured records like sessions or temporary user data without serializing to JSON strings.

Implements Redis list operations through MCP tools supporting append (LPUSH/RPUSH), removal (LPOP/RPOP), and range retrieval (LRANGE). Lists maintain insertion order and support blocking operations, enabling use cases like task queues, activity logs, and sliding windows. Agents can express intent like 'add to queue' which maps to RPUSH, or 'get recent items' which maps to LRANGE with negative indices for tail elements.

Provides Redis set operations through MCP tools for managing unordered collections of unique elements. Tools support SADD (add members), SREM (remove members), SMEMBERS (retrieve all), SISMEMBER (test membership), and set algebra operations (SUNION, SINTER, SDIFF). Sets automatically deduplicate elements and support fast membership testing (O(1) average case), enabling use cases like tag management, user groups, and deduplication.

Implements Redis sorted set operations through MCP tools for managing ranked or scored collections. Tools support ZADD (add scored members), ZRANGE (retrieve by rank), ZRANGEBYSCORE (retrieve by score range), and ZREM (remove members). Sorted sets maintain elements in score order, enabling leaderboards, priority queues, and time-series data with automatic ordering. Agents can express intent like 'get top 10 items' which maps to ZRANGE with WITHSCORES.

Provides Redis stream operations through MCP tools for building append-only event logs with consumer group support. Streams maintain insertion order, assign unique IDs, and support XADD (append), XRANGE (retrieve by ID range), and consumer group operations. Agents can express intent like 'store entire conversation in stream' which maps to XADD, enabling immutable event logs for audit trails, message queues, and conversation history without explicit ID management.

Implements Redis Pub/Sub operations through MCP tools for publish-subscribe messaging patterns. Tools support PUBLISH (send message to channel) and subscription management, enabling real-time message distribution to multiple subscribers. Agents can express intent like 'broadcast notification' which maps to PUBLISH, enabling event-driven architectures where agents publish events and other agents subscribe to channels without explicit message queuing.