Zektor

What Zektor does

The Zektor MCP server connects an MCP-compatible editor or agent to Zektor.io, which provides managed PostgreSQL databases and Valkey caches hosted in Germany. It turns the service’s instance-management API into MCP tools for reading account and resource information, creating instances, changing plans, managing PostgreSQL storage, and deleting instances.

The server also supports a no-account trial flow. When no token is configured, an agent can request a temporary PostgreSQL database that lasts for one hour and includes a claim link. The trial has limits on storage and connections, and the data is removed if it is not claimed within the stated period.

This belongs in the Databases category for workflows where an agent needs to provision or operate hosted database infrastructure. It is not presented as a general-purpose SQL or cache query interface.

How it works

Run zektor mcp to start the MCP server over standard input and output. The client starts the command and communicates with it using MCP. The server uses credentials already managed by the Zektor CLI, so a separate MCP-specific setup is not required after login.

Most provisioning and scaling operations run in the background. The corresponding tools return an action identifier, which can be passed to get_action to inspect the result. Clients can call list_plans and list_regions before creating an instance instead of guessing plan or region names.

Destructive and billable operations require care. Creating an instance, changing its plan, resizing PostgreSQL storage, and deleting an instance can affect cost or data. Deleting is disabled by default in the MCP tool set, according to the README’s guardrail design.

Setup and configuration

Install the CLI with Node.js 18 or newer, then run it through an MCP client using a configuration such as:

{
  "mcpServers": {
    "zektor": { "command": "npx", "args": ["-y", "zektor", "mcp"] }
  }
}

Authenticate with zektor login after creating an access token in the Zektor dashboard. The credential is stored locally in the CLI configuration with restrictive file permissions. For automation, ZEKTOR_TOKEN can supply the token and takes precedence over the stored configuration. Claude Code can register the server with claude mcp add zektor -- npx -y zektor mcp.

Tools and capabilities

The Zektor MCP server provides tools for:

• Reading the current account, listing instances, and retrieving one instance.
• Listing available plans and regions.
• Creating a managed database or cache and checking its background action.
• Scaling an instance to another plan.
• Retrieving connection details; cache connection retrieval returns a live password.
• Viewing PostgreSQL storage, increasing its size, and enabling or disabling automatic growth.
• Deleting an instance when the destructive operation is enabled.
• Creating a temporary trial PostgreSQL database without a configured token.

Storage resizing applies to PostgreSQL and is irreversible: storage can grow but cannot shrink. Storage above the plan’s included amount is billed, and automatic growth also requires a maximum limit.

Limitations and notes

The project describes both the CLI and MCP server as beta software. PostgreSQL connection credentials are not fetched in the same way as cache credentials; the CLI notes that a PostgreSQL role password is shown when created or rotated and is not stored by the server.

Plan changes, storage growth, and instance creation can change the bill. Scaling down can reduce available memory while retaining existing storage, which may produce additional storage charges. Legacy databases created before growable storage are not yet eligible for storage growth.

The trial database is temporary and constrained to one hour, 200 MB, three connections, and three trials per network per day. The README identifies standard input/output as the MCP transport and specifically documents Claude Code registration; it does not document other client-specific setup.