Oracle published a custom Model Context Protocol (MCP) server for Oracle Cloud Infrastructure (OCI). The server lets users manage compute, storage, and networking through natural-language queries instead of CLI or console commands.
Key facts
- Oracle published first-party MCP server for OCI cloud management.
- Supports compute, storage, networking via natural-language queries.
- Wraps OCI Python SDK and CLI for authenticated API calls.
- MCP ecosystem crossed 9,400 servers as of May 2026.
- AWS, GCP, Azure have not released equivalent first-party MCP servers.
Oracle published a custom Model Context Protocol (MCP) server for Oracle Cloud Infrastructure (OCI), according to an Oracle Blogs post. The server exposes OCI resources — compute instances, block volumes, virtual cloud networks, object storage buckets — as MCP tools. Users can run natural-language queries like 'list all running compute instances in us-ashburn-1' or 'create a 100GB block volume named analytics-data'.
The MCP server wraps the OCI Python SDK and the OCI CLI, translating tool calls into authenticated API requests. It supports both synchronous and asynchronous operations, with the ability to poll for long-running tasks such as instance creation. The implementation follows Anthropic's MCP specification (v1.0, released November 2024) and can be used with any MCP-compatible client — Claude Desktop, Claude Code, or third-party hosts.
Unique take: This is the first major cloud provider to ship a first-party MCP server for its own infrastructure. AWS, Google Cloud, and Azure have not published equivalent MCP servers as of this writing. Oracle's move signals that cloud vendors see MCP as a viable alternative to proprietary agent frameworks (e.g., Amazon Bedrock Agents, Google Vertex AI Agent Builder). By adopting an open protocol, Oracle reduces lock-in risk for customers who want to manage multi-cloud environments through a single AI interface.
Implementation details
The server uses the OCI Python SDK for resource discovery and the OCI CLI for authentication. Tool definitions are auto-generated from OCI API schemas. Each tool maps to a specific OCI API endpoint — for example, list_compute_instances calls oci.core.compute_client.list_instances. The server handles pagination and error states, returning structured JSON responses that the LLM can parse. Oracle recommends deploying the server on a secure bastion host with IAM policies scoped to read-only or write operations depending on the use case.
Context: MCP ecosystem growth
MCP crossed 9,400 servers in May 2026, per our previous reporting. Oracle's entry is notable because it comes from a cloud provider, not a third-party developer. The protocol's open nature allows Oracle to maintain the server independently while benefiting from the broader MCP client ecosystem. Anthropic's Claude Code and Claude Desktop are the primary clients, but OpenAI and Google have not yet adopted MCP natively — they rely on proprietary tool-use APIs.
Oracle has not disclosed any internal adoption metrics or latency benchmarks for the MCP server. The blog post notes the server is 'available now' via the Oracle Samples GitHub repository under the Universal Permissive License.
Key Takeaways
- Oracle released a custom MCP server for OCI, enabling natural-language cloud management.
- First major cloud provider to ship a first-party MCP server.
What to watch
Watch whether AWS or Google Cloud ship their own MCP servers within 90 days. If they do, MCP becomes the de facto standard for cloud agent tooling. If they don't, Oracle gains a first-mover advantage in MCP-based cloud ops.
