Talk to your PhoenixAI Clusters Right From Claude: Introducing the PhoenixAI MCP Connector

Query your data, manage your clusters, and get cost insights conversationally, without ever leaving Claude Console

We’re excited to announce the PhoenixAI Custom Connector for Claude, now available for PhoenixAI Cloud customers. With a one-time setup, your team can connect Claude directly to PhoenixAI and start asking real questions of real data — in plain English, with results streaming back into the chat.

If Agent Fawkes brought AI inside the Cloud console, the Claude connector takes the next step: it brings PhoenixAI into the AI assistant your team already lives in.

PhoenixAI is the fastest SQL engine for real-time and AI-driven analytics on the lakehouse. Trusted by Pinterest, Tencent, Expedia, and more.

Why a Claude connector — and why now

Modern data teams don’t work in one tool anymore. They jump between the SQL editor, the BYOC console, the docs, the cost dashboard, and increasingly — Claude. Every context switch is friction. Every copy-paste between tools is a chance to drop a filter, miss a partition, or miss the right answer entirely.

What the connector can do today

Once an admin enables the connector and a user authenticates via OAuth, Claude gets a curated toolkit of PhoenixAI actions. Ask in plain English, get real answers from your live cluster.

Query and explore your data

• Run SQL — read-only or read-write — with results returned as structured, citable output.
• Explore tables with full PhoenixAI-aware metadata: column types, partitions, primary keys, distribution keys, sort keys.
• Sample rows to quickly understand the shape of any dataset before writing the real query.
• Search and browse databases, tables, and schemas conversationally.

Understand and control cost

• Estimate query cost before you run it, using EXPLAIN output and live cluster pricing.
• Pull CCU consumption and USD spend by cluster, warehouse, or time period.
• Find the most expensive queries in your workload — by CCU or runtime — over any window.
• Get optimization recommendations — right-sizing, idle suspension, query rewrites, materialized view candidates, partition pruning opportunities.

Get help, grounded in the docs

• Search PhoenixAI docs by keyword or topic, with clickable citations back to the source.
• Look up SQL syntax — including PhoenixAI-specific extensions — with working examples.

Streaming responses, conversation memory, and the same interactive feel Claude users already expect — now with your data behind it.

Built for the BYOC trust model

The connector was engineered with the same principles that drove BYOC itself: customer data stays in customer infrastructure, admins stay in control, and safety is never traded for convenience.

• OAuth 2.0 with PKCE. No PhoenixAI credentials are ever stored on the Anthropic side. Tokens refresh transparently.
• Per-account MCP server. Each BYOC account runs its own MCP server, scoped to that tenant. Tenant isolation is enforced via JWT validation on every request — no cross-tenant data leakage.
• Full RBAC enforcement. Every tool call inherits the authenticated user’s permissions. The connector cannot escalate privileges; it can only do what the user could already do.
• Admin-controlled enablement. Account admins toggle the MCP server on or off from the BYOC console. OAuth applications are created and revoked from a dedicated settings panel, with named clients and rotatable secrets.
• Confirmation for destructive actions. Scaling a cluster, executing DML, or anything irreversible requires explicit user confirmation in the chat.
• Rate limiting per tool category. Per-user limits — like 100 queries per hour, 10 scale operations per day — prevent runaway agents and protect cluster stability.
• Full audit trail. Enable/disable events, OAuth lifecycle, and every MCP tool invocation flow into the existing PhoenixAI query audit log. Latency, error rates, and token consumption are emitted as metrics.
• TLS 1.2+ everywhere, with optional Anthropic IP allowlisting for customers running behind corporate firewalls.

Who benefits

• Analysts and SQL writers stay in their conversation — ask, query, refine, chart — without juggling five tabs.
• Data engineers and platform teams can check cluster health, surface slow queries, and triage cost anomalies right from chat, with no console hop.
• FinOps and engineering leaders get always-on cost insight: which queries are expensive this week, which warehouses are sitting idle, which clusters need to be right-sized.
• New PhoenixAI users get a guided ramp — schema, syntax, best practices — with the docs and examples cited inline.
• Security and compliance reviewers get a system that was designed around their concerns: OAuth, RBAC, audit logs, tenant isolation, and confirmation gates on destructive actions.

How it works — the 60-second tour

1. PhoenixAI runs an MCP server per BYOC account, acting as a bridge between Claude and your PhoenixAI APIs and SQL endpoints.
2. An account admin enables the MCP server from the BYOC console settings, copies the server URL, and registers Claude as an OAuth application — generating a client ID and client secret.
3. A Claude workspace admin adds the PhoenixAI connector in Claude settings using the URL and OAuth credentials. One-time setup per org.
4. Each user connects via OAuth the first time they use it, granting consent for the specific permissions they need.
5. Once connected, Claude exposes PhoenixAI tools — query, schema, cluster, cost, docs — as first-class actions in the conversation. Ask a question, Claude picks the right tool, the MCP server executes it, results come back streamed.

What’s next on the roadmap

The v1 connector covers the core surface area — query, cluster, schema, cost, docs. Coming next:

• Materialized view management — create, refresh, and manage MVs through conversational commands.
• Data ingestion monitoring — CDC pipelines, streaming ingestion, and routine load status surfaced in chat.
• AI-powered query optimization — Claude reasoning over slow queries to rewrite, suggest indexes, and propose better table designs.
• Multi-cluster orchestration — cross-cluster queries and workload-balancing recommendations.
• Data catalog integration — semantic layer with business glossary, column descriptions, and lineage.
• Proactive alerts — cluster health, cost spikes, and query performance anomalies pushed into Claude rather than waiting to be asked.

Get Started Today

The PhoenixAI Claude connector is rolling out now to PhoenixAI Cloud BYOC customers, at no additional cost during the preview period.

1. Visit the PhoenixAI Cloud BYOC console at https://cloud.PhoenixAI.com/
2. Have your account admin enable the MCP server from Account → Settings, and create an OAuth application for Claude.
3. Add the PhoenixAI connector in Claude settings using the server URL and OAuth credentials.
4. Authenticate, and start asking your data questions — in the assistant your team already uses.

We’d love your feedback as you put the connector through its paces. Every conversation is an opportunity to make the next version sharper — tell us what worked, what surprised you, and what you wish Claude could do next with your PhoenixAI clusters.

---