Tecton

You can run Tecton’s MCP server locally and connect it to MCP clients like Cursor, Claude Code, or Augment to accelerate feature engineering with tool-based guidance. This server runs as a subprocess spawned by your MCP client, providing access to Tecton-specific capabilities and live SDK/reference data to help you build features faster.

How to use

Use the MCP server to enable intelligent assistance inside your editing environment. You connect your MCP client to the local server, then you can query Tecton tools for code examples, SDK references, metric data, and more. Your client will spawn the MCP server as needed, so you don’t manage a separate background service. Start by ensuring the server is reachable from your editor, then begin asking questions like how batch and real-time feature views differ, or request the SDK reference for a specific class.

How to install

Prerequisites: you need the uv package manager available on your system.

brew install uv

Next, verify your setup by running a quick smoke test in your local clone. This confirms the MCP server can start and that Cursor can spawn it when needed.

MCP_SMOKE_TEST=1 uv --directory <path-to-local-clone> run mcp run src/tecton_mcp/mcp_server/server.py

Log into your Tecton cluster to authenticate the SDK used by the MCP server.

tecton login yourcluster.tecton.ai

Configure MCP server connections for clients

You can connect your MCP client to the local MCP server using stdio configuration. The following configurations are shown for common clients to wire up the local server.

{
  "mcpServers": {
    "tecton_cursor": {
      "command": "uv",
      "args": [
        "--directory",
        "<path-to-local-clone>",
        "run",
        "mcp",
        "run",
        "src/tecton_mcp/mcp_server/server.py"
      ]
    },
    "tecton_claude": {
      "command": "uv",
      "args": [
        "--directory",
        "<path-to-local-clone>",
        "run",
        "mcp",
        "run",
        "src/tecton_mcp/mcp_server/server.py"
      ]
    }
  }
}

How to install with Cursor, Claude Code, or Augment

You can wire the MCP server to Cursor, Claude Code, or Augment with the exact commands shown in their configuration flows.

{
  "mcpServers": {
    "tecton_cursor": {
      "command": "uv",
      "args": [
        "--directory",
        "<path-to-local-clone>",
        "run",
        "mcp",
        "run",
        "src/tecton_mcp/mcp_server/server.py"
      ]
    }
  }
}

Troubleshooting

If your MCP integration isn’t picking up tools, restart the editor or reload the MCP extension window so new tools register correctly.

Additional configuration notes

For best results, symlink the client rules into your feature repository so updates propagate automatically.

If you are using a JSON-based setup, place the server configuration in your editor’s MCP settings exactly as shown in the examples above.

Security and access

Authenticate your SDK before connecting to the MCP server to ensure secure access to your Tecton environment and feature services.

Notes and tips

The MCP server runs as a subprocess spawned by the client. You do not need to start or manage a persistent background service manually. The server will be loaded automatically when needed.

Implementation details

The MCP server exposes a set of tools for code exploration and SDK/reference data. Tools include code snippet search, documentation indexing, and SDK references, plus metrics access when an API key is configured.

Available tools

query_example_code_snippet_index_tool

Finds relevant Tecton code examples using a vector database to show usage patterns before writing new code.

query_documentation_index_tool

Retrieves Tecton documentation snippets based on a query to provide context from official docs.

get_full_tecton_sdk_reference_tool

Fetches the complete Tecton SDK reference including all available classes and functions for a broad SDK overview.

query_tecton_sdk_reference_tool

Fetches the Tecton SDK reference for a specified list of classes or functions for targeted information.

query_tecton_metrics_tool

Queries the Tecton Metrics API to return system metrics in human-readable or OpenMetrics format.