Oracle Database MCP Server

Oracle ships several MCP server offerings under one umbrella. The Autonomous AI Database MCP Server is a fully managed, multi-tenant endpoint that ships with every Oracle Autonomous AI Database (versions 19c and 26ai). It exposes Select AI Agent tools (both built-in and custom) to any MCP-compatible client (Claude Desktop, VS Code Cline, OCI AI Agent) with no customer-side infrastructure to operate. Access is governed by database identity, authorization, and Oracle's minimum-privilege policy framework.

The SQLcl MCP Server is built into Oracle SQLcl 25.2+ and runs locally. It abstracts SQLcl commands into MCP tools so an AI client can list saved connections, connect, run SQL/PLSQL, and invoke SQLcl-native commands against any Oracle Database (on-prem, Autonomous, or 23ai Free). All LLM-generated queries are tagged with /* LLM in use ... */ comments and recorded in DBTOOLS$MCP_LOG for auditing.

The github.com/oracle/mcp repository hosts reference implementations for OCI services (oci-api-mcp-server, oci-cloud-mcp-server, dbtools-mcp-server and others), distributed as Python packages installable via uvx and authenticated through the OCI CLI or OCI IAM confidential applications. Oracle labels these as proof-of-concept/reference implementations rather than production-grade servers.

Tools

There are three distinct setup paths depending on which Oracle MCP server you use.

1. SQLcl MCP Server (local)

• Download SQLcl 25.2 or later: https://download.oracle.com/otn_software/java/sqldeveloper/sqlcl-latest.zip
• Save at least one Oracle Database connection in SQLcl (connect -save myconn ...).
• Configure your MCP client to launch SQLcl in MCP mode:

{
  "mcpServers": {
    "sqlcl": {
      "command": "/path/to/sqlcl/bin/sql",
      "args": ["-mcp"]
    }
  }
}

The Oracle SQL Developer extension for VS Code (25.2+) auto-registers the SQLcl MCP server for GitHub Copilot.

2. Autonomous AI Database MCP Server (managed)

• Provision an Autonomous AI Database (19c or 26ai).
• Enable the MCP Server feature and register Select AI Agent tools.
• Point your MCP client at the per-database MCP endpoint using OAuth2/OCI IAM credentials. See https://docs.oracle.com/en-us/iaas/autonomous-database-serverless/doc/mcp-server.html.

3. OCI Reference Servers (from oracle/mcp)

Prerequisites: uv, Python 3.13, OCI CLI, and oci session authenticate --region=<region> --tenancy-name=<tenancy>.

{
  "mcpServers": {
    "oracle-oci-api-mcp-server": {
      "type": "stdio",
      "command": "uvx",
      "args": ["oracle.oci-api-mcp-server@latest"],
      "env": {
        "OCI_CONFIG_PROFILE": "<profile_name>",
        "FASTMCP_LOG_LEVEL": "ERROR"
      }
    }
  }
}

For HTTP transport, set IDCS_DOMAIN, IDCS_CLIENT_ID, IDCS_CLIENT_SECRET, IDCS_AUDIENCE, ORACLE_MCP_BASE_URL, ORACLE_MCP_HOST, ORACLE_MCP_PORT, and OCI_REGION, then connect as:

{
  "mcpServers": {
    "oracle-oci-cloud-mcp-server": {
      "type": "streamableHttp",
      "url": "http://127.0.0.1:8888/mcp"
    }
  }
}

• Run natural-language SQL exploration against an Oracle Database from Claude Desktop or VS Code via the SQLcl MCP Server, including schema introspection and PL/SQL execution.
• Let AI agents call governed Select AI Agent tools on an Autonomous AI Database without exposing raw SQL or maintaining custom infrastructure.
• Automate routine Oracle DBA tasks (DDL generation, Liquibase changelogs, performance diagnostics) through run-sqlcl.
• Provide an AI agent with OCI tenancy operations (compute, identity, database tools) using OCI CLI credentials through the oracle/mcp reference servers.
• Build multi-agent workflows that combine database queries with OCI API actions, all gated by Oracle IAM and audited via DBTOOLS$MCP_LOG.

• "List my saved Oracle connections, connect to PROD_READONLY, and show me the top 10 longest-running queries from V$SQL_MONITOR."
• "Using SQLcl, generate a Liquibase changelog for the HR schema."
• "Through the Autonomous AI Database MCP, run the Select AI Agent tool that summarizes Q3 sales by region."
• "Use the OCI API MCP server to list all compute instances in us-ashburn-1 that are stopped."
• "Create a new Database Tools connection in OCI for the FINANCE Autonomous Database."

• Official, vendor-maintained MCP coverage across managed (Autonomous AI DB), local (SQLcl), and cloud (OCI) surfaces.
• Strong governance: minimum-privilege enforcement, query tagging, and audit logging in DBTOOLS$MCP_LOG.
• SQLcl MCP Server requires no extra install once SQLcl 25.2+ is on the machine.
• Managed Autonomous AI Database MCP removes the need to run server infrastructure.

• The oracle/mcp GitHub repo is explicitly labeled reference/proof-of-concept, not production-ready.
• Autonomous AI Database MCP Server is limited to Oracle Autonomous AI Database 19c and 26ai customers.
• Documentation is fragmented across the GitHub repo, SQLcl docs, and OCI docs, with no single tools catalog.

• MCP Toolbox for Databases (Google) supports Oracle alongside other engines.
• Community Oracle MCP servers such as mcp-server-oracle on PyPI for plain SQL access without SQLcl.
• Generic JDBC/ODBC MCP servers when only basic read access is needed.