BetterDB MCP Server: Ask Your Assistant About Your Valkey and Redis Databases

• Lists the Valkey/Redis servers BetterDB watches

  Shows every database registered in BetterDB with its connection status and capabilities. It is how the assistant finds the instance to look at.

  Try: “Which Valkey instances does BetterDB monitor, and are they connected?”

  Read · list_instances

• Chooses which instance later calls use

  Sets the active instance for the session so the other tools do not need an instance id each time. It changes only the assistant's own selection.

  Try: “Switch to my production Valkey instance.”

  Read · select_instance

• Registers a new database in BetterDB

  Sends a host, port, username and password for a Valkey or Redis server to BetterDB so it starts monitoring it, and can make it the default. The password passes through the chat.

  Try: “Add my staging Valkey at cache.internal:6379 and make it the default.”

  Write · add_connection

• Checks database credentials without saving them

  Tries to connect to a Valkey or Redis server with the details you give and reports whether it worked. Nothing is stored.

  Try: “Test whether I can connect to cache.internal on port 6379.”

  Network · test_connection

• Removes a database from BetterDB

  Deletes a registered connection from BetterDB so it is no longer monitored. It does not delete data on the database itself.

  Try: “Remove the old staging connection from BetterDB.”

  Delete · remove_connection

• Sets the default database

  Marks one registered connection as the default that BetterDB uses when no instance is named.

  Try: “Make the production connection the default.”

  Write · set_default_connection

• Reads the full INFO stats

  Returns the INFO output for the active instance: memory, clients, replication, keyspace, hit rate and ops per second. You can limit it to one section such as memory or stats.

  Try: “Show me the memory section of INFO for my cache.”

  Read · get_info

• Lists the latest slow commands

  Reads the most recent entries from the live slowlog, the commands that took longer than the server's slow threshold.

  Try: “What were the last 20 slow commands?”

  Read · get_slowlog

• Lists recent COMMANDLOG entries

  Reads the newest COMMANDLOG entries, which also cover large requests and replies. Works on Valkey 8 and later only.

  Try: “Show recent COMMANDLOG entries for large replies.”

  Read · get_commandlog

• Reads latency event history

  Returns the server's recorded latency events so you can see which kinds of delay happened and when.

  Try: “Has my server logged any latency events?”

  Read · get_latency

• Runs a memory diagnosis

  Returns the MEMORY DOCTOR verdict and the MEMORY STATS breakdown for the active instance.

  Try: “Is there a memory problem on this instance?”

  Read · get_memory

• Lists connected clients

  Returns the active client list with connection details such as address, age and last command.

  Try: “Which clients are connected right now?”

  Read · get_clients

• Gives a one-call health summary

  Summarizes keyspace hit rate, memory fragmentation, connected clients, replication lag on replicas and keyspace size. Its description tells the assistant to call it first in any investigation.

  Try: “How healthy is my production Valkey?”

  Read · get_health

• Groups slow commands into patterns

  Reads stored slowlog data and groups commands by normalized pattern, with how often each ran and the average time. The history survives the server's own slowlog rotation.

  Try: “Which command patterns are slowest over the past day?”

  Read · get_slowlog_patterns

• Reads stored COMMANDLOG entries

  Returns persisted COMMANDLOG entries and lets you filter by time range to look into a specific incident. Valkey 8 and later only.

  Try: “Show COMMANDLOG entries between 14:00 and 15:00 yesterday.”

  Read · get_commandlog_history

• Groups COMMANDLOG entries into patterns

  Like the slowlog patterns tool, but it also includes large-request and large-reply patterns. Valkey 8 and later only.

  Try: “Which commands send the largest replies?”

  Read · get_commandlog_patterns

• Lists detected anomalies

  Returns events that BetterDB's Z-score checks flagged on memory, hit rate, CPU and other metrics. Needs a BetterDB license tier that includes it.

  Try: “Were there any anomalies in the last 24 hours?”

  Read · get_anomalies

• Shows client activity over time

  Returns time-bucketed snapshots of connection counts, command mix and buffer use, with start and end times to focus on one window.

  Try: “How did client connections change during last night's incident?”

  Read · get_client_activity

• Finds the busiest keys

  Returns hot keys from stored scans, using LFU frequency scores when the eviction policy is an LFU one, or idle time and COMMANDLOG data otherwise.

  Try: “Are any hot keys causing uneven load?”

  Read · get_hot_keys

• Lists cluster nodes

  Finds every node in a Valkey cluster with its role, address, health and slot ranges. Returns a message if the instance is not a cluster.

  Try: “List the nodes in my cluster and their slot ranges.”

  Read · get_cluster_nodes

• Compares node performance

  Returns memory, ops per second, clients, replication offset and CPU for each cluster node, to spot hot nodes or lagging replicas.

  Try: “Which cluster node has the highest memory usage?”

  Read · get_cluster_node_stats

• Merges slowlogs across the cluster

  Combines the slowlogs of all nodes into one list, because a single node's slowlog misses the rest of the cluster.

  Try: “Show the slowest commands across the whole cluster.”

  Read · get_cluster_slowlog

• Shows per-slot key counts and CPU

  Returns key counts and CPU time per hash slot, ordered by either one. Valkey 8.0 and later only.

  Try: “Which slots use the most CPU?”

  Read · get_slot_stats

• Shows one latency event over time

  Returns the full history of one named latency event such as command or fast-command. Run the latency tool first to see the event names.

  Try: “Show the history of the fast-command latency event.”

  Read · get_latency_history

• Reads the ACL audit log

  Returns stored ACL audit entries that you can filter by user, reason (auth, command, key, channel) or time, to see why a connection fails.

  Try: “Show failed authentication attempts from the last hour.”

  Read · get_acl_audit

• Starts a local BetterDB monitor

  Starts BetterDB Monitor as a background process on your computer by running the @betterdb/monitor package with npx, then points the MCP server at it. It keeps running until stop_monitor.

  Try: “Start a local BetterDB monitor on port 3001.”

  Runs code · start_monitor

• Stops the background monitor

  Stops a monitor that was started by start_monitor or by the autostart option.

  Try: “Stop the local BetterDB monitor.”

  Runs code · stop_monitor

• Lists registered caches

  Lists the semantic caches and agent caches registered on the active instance, with hit rate and total operations.

  Try: “Which AI caches are on this instance and how are they doing?”

  Read · cache_list

• Reports one cache's health

  Gives detailed health for one cache. For a semantic cache it breaks results down by category, and for an agent cache by tool.

  Try: “How healthy is the faq-cache?”

  Read · cache_health

• Suggests a similarity threshold

  Looks at the rolling window of similarity scores for a semantic cache and recommends a threshold. It returns an error for agent caches.

  Try: “What similarity threshold should faq-cache use?”

  Read · cache_threshold_recommendation

• Rates each tool's caching

  Shows hit rate, cost saved and a TTL recommendation for each tool in an agent cache. It returns an error for semantic caches.

  Try: “Why is the weather_lookup tool's hit rate so low?”

  Read · cache_tool_effectiveness

• Shows similarity scores as a histogram

  Returns 20 buckets of width 0.1 for recent similarity scores in a semantic cache.

  Try: “Show the similarity score distribution for faq-cache.”

  Read · cache_similarity_distribution

• Lists recent change proposals for a cache

  Shows the latest proposals for one cache, in any status, so the assistant does not propose something already pending or applied.

  Try: “What changes were recently proposed for faq-cache?”

  Read · cache_recent_changes

• Proposes a new similarity threshold

  Creates a pending proposal to change a semantic cache's threshold, globally or per category. It needs a reason of at least 20 characters and waits for approval.

  Try: “Propose lowering the faq-cache threshold and explain why.”

  Write · cache_propose_threshold_adjust

• Proposes a new TTL for a tool

  Creates a pending proposal to change one tool's time to live in an agent cache. It needs a reason of at least 20 characters.

  Try: “Propose a longer TTL for the weather_lookup tool.”

  Write · cache_propose_tool_ttl_adjust

• Proposes clearing cache entries

  Creates a pending proposal to invalidate entries, by search filter for semantic caches or by tool, key prefix or session for agent caches. It warns when more than 10,000 entries would be affected.

  Try: “Propose invalidating all cached entries for the session abc123.”

  Write · cache_propose_invalidate

• Lists proposals waiting for approval

  Shows pending cache proposals on the active instance, newest first, optionally for one cache.

  Try: “What cache proposals are waiting for review?”

  Read · cache_list_pending_proposals

• Reads one proposal and its history

  Returns a single proposal by id with its audit trail.

  Try: “Show proposal prop_123 and who touched it.”

  Read · cache_get_proposal

• Approves a proposal and applies it

  Approves a pending proposal and applies the change to Valkey right away: a new threshold, a new TTL or an invalidation that removes cache entries. Calling it again on an applied proposal returns the earlier result.

  Try: “Approve proposal prop_123.”

  Delete · cache_approve_proposal

• Rejects a pending proposal

  Marks a pending proposal as rejected and can record a reason in the audit trail. Nothing is changed in Valkey.

  Try: “Reject proposal prop_123 because the TTL is too short.”

  Write · cache_reject_proposal

• Edits a proposal and approves it

  Changes the threshold or TTL of a pending proposal and approves it in one step, which applies it to Valkey. Invalidate proposals cannot be edited.

  Try: “Set the threshold in prop_123 to 0.12 and approve it.”

  Write · cache_edit_and_approve_proposal

• Lists agent-memory stores

  Shows the agent-memory stores found on an instance, with their names and capabilities.

  Try: “Which agent memory stores exist on this instance?”

  Read · memory_stores

• Lists memories in a store

  Returns memories from a store, newest first, filtered by scope or tags.

  Try: “Show the 20 newest memories in the support-agent store.”

  Read · memory_list

• Reads one memory

  Fetches a single memory from a store by its id.

  Try: “Show memory mem_42 from the support-agent store.”

  Read · memory_get

• Shows a memory store's size

  Returns the item count, eviction count and live configuration of a memory store.

  Try: “How many memories does the support-agent store hold?”

  Read · memory_stats

• Finds memories similar to a vector

  Searches a store for memories near an embedding you supply, with a result limit and a maximum distance.

  Try: “Recall the 5 closest memories to this embedding.”

  Read · memory_recall

• Proposes deleting memories

  Creates a pending proposal to forget one memory or a scope or tag group. Nothing is deleted until someone approves it.

  Try: “Propose forgetting all memories tagged outdated.”

  Write · memory_forget

• Lists delete proposals waiting

  Shows pending forget proposals, optionally for one store.

  Try: “Which forget proposals are pending?”

  Read · memory_list_pending_forgets

• Approves a delete and applies it

  Approves a pending forget proposal and deletes the memories from the live store.

  Try: “Approve forget proposal fp_7.”

  Delete · memory_approve_forget

• Rejects a delete proposal

  Rejects a pending forget proposal so nothing is deleted.

  Try: “Reject forget proposal fp_7.”

  Write · memory_reject_forget

• Lists AI components found on the database

  Shows semantic caches, agent caches, memory stores and retrieval pipelines BetterDB found, with liveness and the latest metrics.

  Try: “Which AI components are running on this Valkey?”

  Read · ai_list_instances

• Shows an AI component's metrics over time

  Returns stored time series for one component: hits, misses, hit rate, cost saved, evictions, item count, index size and threshold.

  Try: “Plot the hit rate history of faq-cache this week.”

  Read · ai_instance_history

• Lists recent AI traces

  Lists traces ingested through OpenTelemetry, covering LLM calls, cache lookups, memory recalls and retrieval spans. They come from your instrumented apps, not one database.

  Try: “Show the latest AI traces with cache misses.”

  Read · list_ai_traces

• Reads one trace in full

  Returns every span of a trace with timing, parent links and attributes such as model, cache hit or miss and similarity score.

  Try: “Show the span waterfall for trace abc.”

  Read · get_ai_trace

• Explains why a trace missed

  Joins a trace's cache and memory spans with live Valkey state: whether the key still exists, its TTL, the active threshold and the index state.

  Try: “Why did trace abc miss the cache?”

  Read · correlate_ai_trace

• Forecasts when a limit is reached

  Projects one metric (ops per second, used memory, CPU or memory fragmentation) forward and estimates the time until it hits its ceiling.

  Try: “When will used memory reach its limit?”

  Read · get_forecast

• Lists sustained latency regressions

  Returns detected cases where p99 command latency stayed worse than baseline, from the same store as the anomaly tool. Needs a license tier that includes it.

  Try: “Any latency regressions this week?”

  Read · get_latency_regressions

• Lists the biggest keys

  Ranks the largest keys by measured memory from key analytics snapshots. The snapshots favor collections with many elements. Needs a license tier that includes it.

  Try: “What are the 10 largest keys?”

  Read · get_largest_keys

• Checks vector search indexes

  Shows each search index with document count, memory use, indexing failures and percent indexed. It needs the Search module on the database.

  Try: “Are my vector indexes fully indexed?”

  Read · get_vector_indexes

• Shows search latency per index

  Returns p50, p95 and p99 FT.SEARCH latency for each vector index, plus SLA breach status when the license tier supports it.

  Try: “What is the p99 search latency of my vector indexes?”

  Read · get_inference_latency

Watch out for

• The assistant can approve its own changes

  The propose tools say a human must approve, but the approve tools are part of the same server. An assistant that can call them can apply cache changes and delete memories itself. Use a token without admin rights, or tell the assistant never to approve. npmjs.com

• It can delete memories and cache entries

  Approving a forget proposal deletes memories from the live store. Approving an invalidate proposal removes cache entries in Valkey. remove_connection also removes a monitored database from BetterDB. npmjs.com

• Autostart runs a second package on your machine

  The --autostart option and start_monitor run npx --yes @betterdb/monitor in the background, with your environment variables passed on. Skip autostart if you do not want that, and point BETTERDB_URL at a monitor you already trust. npmjs.com

• Database passwords pass through the chat

  add_connection and test_connection take a database host and password as tool inputs, so the password appears in your conversation. Add connections in the BetterDB settings instead. npmjs.com

• Tool usage is reported to your BetterDB instance

  The server sends each tool name, success, duration and error to your BetterDB instance. Set BETTERDB_TELEMETRY=false to turn this off. npmjs.com

Security advisories