If you use Oracle to run your business, your most important data already lives there. The question is how to let Claude work with that data without weakening the controls around it. You can set this up in five steps, from prerequisites to a verified first prompt, with CData Connect AI, a managed Model Context Protocol (MCP) platform that connects AI assistants like Claude to live enterprise data sources.
Why connect Oracle to Claude with MCP?
MCP is an open standard introduced by Anthropic and now governed by the Agentic AI Foundation under the Linux Foundation that connects AI assistants to the systems where data lives. The model reasons and the MCP server handles access, putting authentication, permissions, and logging in a layer your team controls. Connecting Oracle to Claude is a governance decision: who Claude acts as, what it can reach, and how you prove it later. One protocol replaces a custom pipeline for every AI-to-data pairing. Oracle includes an MCP server in SQLcl, usable with any MCP-capable client, including Claude Desktop, for local developer work.
SQLcl covers an individual desktop. A team needs a hosted endpoint, because Claude connects to remote MCP servers from Anthropic's cloud, and a security review needs central identity and logging. Anthropic selected Connect AI as its certified partner for enterprise data connectivity. One endpoint connects Claude chat, Claude Desktop, Claude Code, and the Anthropic Agent SDK to hundreds of systems, including Oracle. Identity passthrough enforces your existing Oracle permissions at runtime, and data stays in Oracle.
Prerequisites for connecting Oracle to Claude
Have these three things ready before you start:
Oracle database access: A running Oracle Database that is reachable over the network, plus a database user and password with read access to the data you want Claude to work with. Keep the host, port, and service name (or SID) at hand. For Oracle Autonomous Database, download the wallet file (cwallet.sso) as well.
Claude plan with connector support: Custom connectors using remote MCP are available on all plans, from Free through Enterprise, with Free limited to one custom connector
CData Connect AI account: Sign up for a free trial, which is enough to complete every step in this guide. You install nothing locally. Connect AI manages the connectivity for you.
Step-by-step: how to connect Oracle to Claude
Step 1: Add and configure your Oracle database connection
Log in to Connect AI, click Sources, and then click + Add Connection
Select Oracle, then enter the connection details you gathered. Oracle supports the EZConnect format, which defines a connection as host, port, and service name. For Oracle Autonomous Database, upload the cwallet.sso wallet file for Secure Sockets Layer (SSL) authentication
Click Save & Test and wait for the success message before moving on
Step 2: Set up the CData Connect AI managed MCP endpoint
A remote MCP server is an internet-hosted MCP server that Claude reaches over the web, with nothing running on your local machine. Claude connects to it from Anthropic's cloud, not from your device, so the endpoint must be reachable over the public internet. Connect AI hosts this endpoint for you. There is nothing to install, patch, or run, and your database stays behind your own network controls.
Step 3: Connect the MCP endpoint to Claude
In Claude, open Settings, click Connectors, then click Browse connectors
Find CData Connect AI and click Connect. Custom connectors using remote MCP work across claude.ai, Claude Desktop, and the mobile apps, so you set this up once and it follows you to every Claude client
Step 4: Authenticate with OAuth/SAML identity passthrough
When you click Connect, Claude opens a sign-in window for your Connect AI account. Sign in with your own credentials, review the requested access, and approve it. That sign-in ties every request Claude makes to your individual identity, which is what makes the audit trail attributable.
Step 5: Verify the live Oracle-to-Claude connection with a test prompt
Start a new chat and confirm the connector is enabled under the tools menu. Begin with read-only validation prompts, then move to business questions. Because of identity passthrough, each prompt returns only what your Oracle permissions allow, so two users running the same prompt get different results based on their own Oracle privileges. For a detailed walkthrough, see our KB article on connecting Claude to Oracle.
Prompt type | Example prompt | What Claude does |
Validation | "List my saved Oracle connections and confirm the connected user, do not fetch data yet." | Reads connection metadata and the active session, no data fetched |
Business question | "Show me Q2 revenue by region from our Oracle ERP." | Fetches and aggregates live data in Oracle |
Business question | "Which customers have overdue invoices above 10,000?" | Fetches matching records across related objects |
Agent building | "Describe the orders object, then draft a request for monthly totals." | Reads the structure, proposes the request for review |
Compare Claude's answers with the same lookups run directly in Oracle. Matching results mean the connection is live and correct. For more ideas, see the CData prompt library, and teams on Oracle applications can follow the parallel paths for Oracle ERP and Oracle HCM.
Enterprise security and governance for Oracle data in Claude
A security review needs two things proven: that Claude cannot fetch more than the requesting user could, and that every request is traceable
Identity-first security and Oracle RBAC passthrough
The Step 4 sign-in uses OAuth, a standard that lets an app act on your behalf without holding your password. Claude connectors use delegated, per-user OAuth, so there are no shared service accounts and tokens are never shared between users. Inside Oracle, Role-Based Access Control (RBAC) groups privileges into roles and enforces object privileges per user. Identity passthrough checks every Claude request against that same rulebook. With identity passthrough, every request Claude makes runs as the requesting user, so it can only read data that user's RBAC permissions already allow.
Audit trails, governance, and no data movement
Connect AI logs every Claude request with full user attribution. On the Oracle side, unified auditing writes audited actions into the read-only UNIFIED_AUDIT_TRAIL view, so once your DBA enables an audit policy covering the connecting user, Claude's activity lands in a permanent log. Requests run in Oracle, so no copy of your data leaves the database.
Enterprise concern | Oracle/CData control | How to prove it in a security review |
User overreach | Per-user OAuth plus Oracle RBAC at request time | UNIFIED_AUDIT_TRAIL shows activity under the individual DBUSERNAME |
Standing elevated access | No shared service accounts | Token scoping in Claude's connector security model |
Tampered evidence | AUD$UNIFIED protected against unauthorized changes | Direct modification attempts fail and are themselves logged |
Data sprawl | Requests run in Oracle, nothing replicated | No extract pipelines or copies to inventory |
Troubleshooting common Oracle-to-Claude connection issues
Oracle connection, JDBC driver, and Oracle SQL errors
Connection refused. Usually a mismatched connection string. EZConnect uses host:port/service_name, the older SID format uses host:port:sid. A wrong service name returns ORA-12514, while a wrong host or port gets no response
Authentication and permission troubleshooting
Claude cannot see the connector. This applies to self-hosted MCP servers, which must be reachable from Anthropic's cloud. If you followed this guide, Connect AI hosts the endpoint for you, so check instead that the connector is enabled in the chat
Connected, but no data returned. Claude inherits only your access, so grant the missing privilege in Oracle and retry the same prompt
Symptom | Likely cause | Fix |
No response on connect | Wrong host or port | Correct them in host:port/service_name |
ORA-12514 | Service name not known to the listener | Use the service name registered with the listener |
Connector not visible in Claude | Blocked ingress | Allow Anthropic's cloud, or use the managed endpoint |
Connection works, no data returned | Missing Oracle privilege | Grant it in Oracle, retry the prompt |
Connect Oracle to Claude with CData Connect AI today
Connecting AI to Oracle clears it as the security review. CData Connect AI clears it as every connection inherits per-user permissions through delegated OAuth, Oracle's unified audit trail records every request, and the same SOC 2 and ISO 27001 audited endpoint connects Claude to hundreds of other data sources your enterprise runs on.
Start your free trial today!
Frequently asked questions
How does the Oracle-to-Claude connection actually work? Does Claude fetch data from my Oracle database directly?
No. Claude connects to an MCP server hosted by Connect AI, which fetches the answer from Oracle under your identity and returns the result. Your database credentials never leave Connect AI.
Will Claude respect my existing Oracle user permissions and security policies?
Yes. Connectors use delegated, per-user OAuth, and identity passthrough enforces Oracle RBAC on every request. If you cannot see a record in Oracle, Claude cannot fetch it for you.
What do I need before I can connect Oracle Database to Claude?
Three things: a reachable Oracle Database with a user that has read access, a Claude plan (all plans support custom connectors), and a Connect AI account. Nothing installs locally.
Why isn't Claude showing my Oracle data or connector after setup?
Confirm the connector is enabled in the current chat, then that the sign-in completed. If Claude connects but returns no data, grant the missing access in Oracle and retry.
Can I use Claude Code or the Anthropic Agent SDK to connect to Oracle, or is this only for Claude Desktop?
Both work. One Connect AI endpoint serves claude.ai, Claude Desktop, mobile, Claude Code, and Agent SDK agents, so you connect Oracle once and every Claude surface gets the same governed access.
