Name: OceanBase
Author: oceanbase

LABSCO SUMMARY

One server per part of the OceanBase stack, not one server that does everything.

Which one you install follows from what you operate: obdiag for reading logs and running inspections when something is wrong, okctl if the clusters live in Kubernetes, obshell for deployment, and the database server itself for plain SQL. They are separate installs, and the diagnostic ones expect their underlying command-line tool to be present and already configured.

What it is

The OceanBase team's repository of MCP servers for their own ecosystem. Rather than one server it publishes eight: the database itself, the OCP cloud platform, OBCloud, okctl for Kubernetes, the obdiag diagnostic tool, obshell for cluster operations, seekdb for vector work, and PowerMem for AI memory.

What you get

OceanBase MCP Server — SQL queries and data management against an OceanBase instance
Obdiag MCP Server for diagnosis: obdiag_check_run returns an inspection report, obdiag_analyze_log finds errors that already occurred, obdiag_gather_log collects logs filtered by time range or grep, and obdiag_cluster_list names the clusters it can address
Okctl MCP Server for OceanBase on Kubernetes — list_all_clusters, show_cluster, scale_cluster, upgrade_cluster, tenant work including switchover_tenant and activate_tenant, backup policies, and execute_cluster_sql for queries
Obshell MCP Server for cluster creation, deployment and operations management
OCP MCP Server for cluster management and monitoring, and OBCloud MCP Server for the cloud service
Seekdb MCP Server — vector operations, collection management, SQL and AI functions for the seekdb database
PowerMem MCP Server for memory storage, search and retrieval with an Ebbinghaus forgetting curve

Requirements

An OceanBase instance to point at; there is a free trial if you do not have one. Each server installs separately, with its own prerequisites. uvx obdiag-mcp takes a mode — stdio, sse or streamable-http, sse on port 8000 by default — and needs obdiag itself installed with at least one config file in ~/.obdiag/, one .yml per cluster. uvx obshell-mcp reads OBSHELL_HOST, OBSHELL_PORT, CLUSTER_NAME and SYS_PASSWORD. The okctl server needs Python 3.10 or higher, the okctl CLI installed, a Kubernetes context with access to your clusters, and OB_CLUSTER_PASSWORD for the sys tenant; --tools narrows what it exposes to any of clusters, tenants, backup_policy, components and sql.

Setup effort

Follow the published steps — git clone https://github.com/oceanbase/awesome-oceanbase-mcp.git

See installation options

USE CASES

01

Finding out why a cluster misbehaved overnight

Run the inspection, then analyse the logs for the errors that already happened.

obdiag_check_runobdiag_analyze_log

02

Collecting evidence for a support thread

Gather the logs bounded by time range and grep, for the cluster you name.

obdiag_cluster_listobdiag_gather_log

03

Scaling a cluster running on Kubernetes

Read the cluster's current shape, then scale it.

show_clusterscale_cluster

04

Querying a cluster without setting up a client

Configure the connection once, then run the SQL through the same server.

configure_cluster_connectionexecute_cluster_sql

05

Promoting a standby tenant

Read the tenant's detail first, then switch primary and standby over.

show_tenantswitchover_tenant

36 tools

Listed in project docs. Your client may show a different list.

obdiag_cluster_list

obdiag_check_run

obdiag_analyze_log

obdiag_display_list

obdiag_display_run

obdiag_gather_log

list_all_clusters

show_cluster

create_cluster

delete_cluster

scale_cluster

update_cluster

upgrade_cluster

list_tenants

create_tenant

BEFORE YOU INSTALL

What to know first

Tools you get

36 tools named in the project's documentation. The 36 combine two of the repository's eight servers, okctl (30) and obdiag (6); no single install gives 36. okctl's --tools option narrows further, for example --tools=clusters exposes 7.

Requires

OB_CLUSTER_PASSWORD in your environment
SYS_PASSWORD in your environment

Where it runs

Both documented clients run the server on your machine.

INSTALLATION

Connect your client

Server setup

Install uv (Python package manager)

obdiag server: install obdiag and put at least one config .yml in ~/.obdiag/, then run uvx obdiag-mcp sse

obshell server: uvx obshell-mcp (optionally with --sse 8000)

TERMINAL

git clone https://github.com/oceanbase/awesome-oceanbase-mcp.git

okctl server: install and configure okctl + a Kubernetes context, then uv run --directory <YOUR_PATH_TO_MCP_OCEANBASE> okctl_mcp_server

MCP Clients

Cursor — obshell

CURSOR — OBSHELL · JSON

{
  "mcpServers": {
    "obshell-mcp": {
      "command": "uvx",
      "args": [
        "obshell-mcp"
      ],
      "env": {
        "OBSHELL_HOST": "127.0.0.1",
        "OBSHELL_PORT": "2886",
        "CLUSTER_NAME": "cluster",
        "SYS_PASSWORD": "<YOUR_SYS_PASSWORD>"
      }
    }
  }
}

Any MCP client — okctl

ANY MCP CLIENT — OKCTL · JSON

{
  "mcpServers": {
    "okctl": {
      "command": "uv",
      "args": [
        "--directory",
        "<YOUR_PATH_TO_MCP_OCEANBASE>",
        "run",
        "okctl_mcp_server"
      ],
      "env": {
        "OB_CLUSTER_PASSWORD": "<YOUR_OB_CLUSTER_PASSWORD>"
      }
    }
  }
}

OceanBase