Local MCP adapter

Configure MCP locally#

1. Choose direct or Docker stdio

   Run the adapter as a local child process. The documented integration is stdio, not a public hosted MCP endpoint.

2. Provide the API origin and key

   Load the production base URL and raw API key from the local client environment or secret configuration.

3. Test capabilities first

   Call modastera_get_capabilities to confirm connectivity and supported workflows before mutating data.

4. Use typed workflow tools

   Manage datasets, annotations, pipelines, results, and deployments with the registered tools and their documented arguments.

5. Inspect the result envelope

   Check success, data, and error information before using returned identifiers in a later tool call.

Trust-boundary rules#

• Allow local path tools only for paths the client and adapter can both access and trust.
• Do not expose the stdio adapter directly to untrusted network clients.
• Require human confirmation for costly compute, deployment changes, destructive edits, or sensitive data operations.
• Call capabilities at runtime instead of relying on a fixed tool count.

MCP setup and tool cards#