AI-native Valkey observability
BetterDB isn't just observable by humans. It gives AI assistants structured access to live and historical database state through MCP - and it observes your AI workloads in return: traces, vector search, semantic caches, and agent memory.
Ask your database questions, get investigated answers
Connect Claude Code, Cursor, or any MCP client to @betterdb/mcp and the assistant can pull real evidence instead of speculating:
Claude / Cursor / AI agent
↓ MCP (60 tools)
BetterDB Monitor
↓
Valkey / Redis60 tools, organized by what agents actually do
Read-mostly by design: investigation is read-only, and the one surface that can change anything runs through an explicit propose-and-approve workflow.
Instance investigation
21 toolsHealth summaries, INFO, slowlog and COMMANDLOG (live and persisted history), latency events, memory doctor, clients, hot keys, ACL audit, cluster nodes and per-slot stats.
Analytics
3 toolsCapacity forecasts with projected time-to-ceiling, sustained P99 latency regressions vs baseline, and largest keys by measured memory.
Vector search
2 toolsPer-index health for valkey-search / RediSearch and FT.SEARCH p50/p95/p99 inference latency with SLA breach status.
AI observability
5 toolsDiscover AI components (semantic caches, agent memory, retrieval), list and inspect OTLP traces, and correlate a trace with live Valkey state to explain a cache miss or stale recall.
Cache intelligence
14 toolsRead cache health and similarity distributions, then propose threshold/TTL changes or invalidations. Proposals are advisory: nothing touches Valkey until a human (or policy) approves.
Connections & lifecycle
6 toolsAdd, test, and switch database connections; start and stop persistent monitor processes.
And in the other direction: observing your AI stack
BetterDB ingests OpenTelemetry traces from AI applications and correlates each span - LLM call, cache lookup, memory recall, retrieval - with the live Valkey state underneath it. Vector search gets per-index health and p50/p95/p99 inference latency; semantic caches get hit-rate health and threshold recommendations with a supervised approval workflow (Pro, free in early access).
FAQ
How can an AI agent monitor Valkey or Redis?
Through MCP (Model Context Protocol). BetterDB ships an MCP server (@betterdb/mcp) exposing 60 tools that let assistants like Claude Code or Cursor query health, slowlogs, COMMANDLOG, latency, memory, clients, hot keys, cluster state, capacity forecasts, and anomalies - over persisted history, not just the live snapshot. The assistant investigates with real data instead of guessing.
Is it safe to give an AI access to my database monitoring?
The MCP surface is read-mostly by design: investigation tools only read metrics and history. The exception is the cache-intelligence workflow, where an agent can propose changes (thresholds, TTLs, invalidations) but nothing is applied to Valkey until a proposal is explicitly approved - every proposal carries an audit trail. Access requires a token created in Monitor settings.
What are AI traces in BetterDB?
BetterDB ingests OpenTelemetry (OTLP) traces from your AI application - LLM calls, cache lookups, memory recalls, retrieval spans - and renders span waterfalls correlated with the live Valkey state underneath each request. That answers questions like "did this slow response come from the model or from a cache miss?"
Does BetterDB monitor vector search workloads?
Yes. For valkey-search and RediSearch it tracks FT.SEARCH ops/sec and latency percentiles per index, per-index health (document count, memory, indexing failures, percent indexed), and SLA breach alerts (SLA features are Pro, free during early access).
Do I need the cloud version to use MCP?
No. The MCP server runs over stdio via npx @betterdb/mcp against any self-hosted BetterDB Monitor, including air-gapped deployments. Create a token under Settings, add the server to your MCP client config, and start asking questions.
Ready to get started?
Start monitoring in minutes - no infrastructure to maintain. Team collaboration, agent-based monitoring for private databases, and more. Or self-host - open source core, zero lock-in.