Skip to main content

MCP Overview

DataStar v3 embeds a Model Context Protocol (MCP) server so an AI assistant (Claude, Kiro, Cline, or any MCP-capable client) can work alongside you, grounded in your actual workspace, database, and backlog.

What the assistant can do​

The server exposes capabilities grouped into a few clear areas:

  • Explore your database. Inspect schema, describe tables, query vendor dictionary and metadata tables, and run read-only SQL against the active connection.
  • Work with components. List, search, and read the components in your workspace, inspect dependencies, and compare scripts against the database or draft versions.
  • Draft and stage changes. Extract schema or data into drafts, review them, apply or discard them, and add scripts to the release basket.
  • Drive deployments. Execute scripts in verify or commit modes and generate deployment packages, all within the permissions you grant.
  • Integrate with your backlog. Work directly with Jira and Azure Boards through a common set of task-tracking tools: query, create, update, transition, and comment on work items without leaving the assistant.

The full tool catalogue lives in the MCP Tool Catalog. The overview deliberately doesn't list every tool, because the shape of the integration matters more than any one name.

How it connects​

The server runs locally alongside the DataStar client:

  • Localhost only. The server binds to 127.0.0.1, and an explicit request-level loopback check rejects anything that is not from the local machine, even if the bind address is ever misconfigured.
  • Port is configurable in preferences (default 3000). It is a per-workspace setting, so multiple DataStar instances can run side-by-side without colliding on a shared port.
  • Authenticated with an API key. Every request must carry an X-API-Key header. The key is generated automatically on first start and can be regenerated at any time from preferences.
  • Read by default, write on request. A separate Allow write access toggle gates any tool that modifies files, drafts, the database, or a work item. Read tools work as soon as the server is running; write tools refuse until you enable them.
  • Scoped to the open workspace. The assistant sees what you see: the workspace currently loaded in DataStar. Close the workspace and the tools return an error.

Configuring a client​

Copy your API key from Preferences → MCP Server and point your MCP client at the local endpoint. A typical client configuration looks like:

{
  "mcpServers": {
    "datastar": {
      "url": "http://127.0.0.1:3000/mcp",
      "headers": {
        "X-API-Key": "your-api-key-here"
      }
    }
  }
}

Worked examples for Claude Desktop, Cline, and Kiro are in Connecting an MCP Client.

The security model​

DataStar treats MCP as a privileged integration and defaults to the safest possible posture:

  • The server is off by default and must be enabled from preferences.
  • Remote connections are refused at both the network layer and the request layer, so only processes on the same machine can reach the server.
  • Connections are rejected without a valid API key, protecting against any other process on the machine.
  • SQL execution is read-only. Queries that try to mutate data are rejected before they reach the database.
  • Destructive or mutating operations are explicit. They are marked as such in the tool schema so compatible clients can prompt before running them.
  • Secrets are never exposed. Connection strings, passwords, and tokens stored by DataStar are not readable through the MCP.

When to use it​

Turn MCP on if you already use an AI assistant day to day and want it grounded in your database, your components, and your work items rather than guessing from general knowledge. Leave it off on shared machines, or where you have not yet decided how AI tooling fits into your workflow; the server stays dormant until you explicitly enable it.

Next steps​

  1. Enable the server and generate your API key.
  2. Configure metadata tables for your platform (optional, but a significant quality-of-life improvement for any system with a dictionary / data-catalogue layer).
  3. Connect your MCP client using the configuration above.
Last updated on