AI Agent

Ask your cluster anything

The built-in agent reads schema and query_log before recommending anything. Connect over MCP from Claude, Cursor or any client — read-only, nothing applied without you.

29tools across schema, diagnostics and optimization
MCPendpoint for Claude, Cursor and any client
BYOMbring your own model and API key
Read-onlynothing applied without you
01 · Grounded answers

It reads your cluster before it speaks

Every recommendation starts from your real schema, table stats and query_log — not generic advice. Ask why a query is slow and the agent inspects the statement, the parts it scans and the indexes it misses.

  • Schema, storage, replication, merges and health tools
  • Recommend-only advisor: skip indexes, projections, PREWHERE rewrites
  • Skill recipes for anything not covered by a primitive tool
02 · MCP server

Bring your own client over MCP

chmonitor exposes the same tools on a Model Context Protocol endpoint at /api/mcp. Point Claude, Cursor or any MCP client at it and query your cluster from wherever you already work.

  • Open, API-key or Clerk OAuth auth postures
  • Same read-only guarantees as the built-in agent
  • Works on Cloudflare Workers, Docker and Kubernetes deploys
03 · Your model, your rules

Bring your own model, skills and servers

Pick the model, plug in your own API key, add skills and connect external MCP servers. The agent adapts to your stack instead of locking you into one vendor.

  • Anthropic, OpenAI, Google and OpenAI-compatible endpoints
  • Per-message stats: tokens, duration, cost estimate
  • Conversation history persisted per user
Gallery

More of AI Agent in action

Capabilities

Everything in the box

Schema tools

Databases, tables, columns, DDL and table stats on demand.

Query diagnostics

Slow, failed and expensive queries with EXPLAIN analysis.

Optimization advisor

Skip indexes, projections and PREWHERE rewrites — recommend-only.

Health & replication

Replication lag, merges, mutations and disk pressure checks.

Visualizations

The agent renders charts and tables inline in the conversation.

Skills

Loadable recipes extend the agent beyond its primitive tools.

FAQ

Common questions

Can the agent change my cluster?

No. The default toolset is read-only; the three destructive control tools ship disabled and must be explicitly enabled by environment variable.

Which models are supported?

Anthropic, OpenAI, Google and any OpenAI-compatible endpoint. Bring your own API key per provider.

Do I need the dashboard open to use MCP?

No. The MCP endpoint is served by the same deployment — connect from Claude, Cursor or any MCP client directly.

Start monitoring — advisor and alerts built in

Open the hosted dashboard and connect a cluster, or self-host with Docker, Kubernetes, or from source. Self-hosting is always free.