5. Follow-along: Define tools
2m

Follow-along: Define tools with local operations (⏳ 10 min)

🎯 Goal: Expose your as an MCP tool.

Step 1: Define a new tool for your agent

  1. In your Sandbox at http://localhost:4000/ you should still have the you designed from the previous step. This query will become your MCP tool!

  2. Add a description above your tool.

    Example description
    # Get the last 5 launches with current weather on the launchpad

    MCP tool descriptions give the agent all the context they need about the tool including , datatypes, and any limitations. When you write a description, you should consider what the agent needs to know to effectively use your tool.

    Find out more about writing tool descriptions in the section below.

  3. Look for other opportunities to give your agent better context through . For example, launchId as a name gives your agent more information than id.

  4. Rename your to be more descriptive, for example GetLastFiveLaunches.

    Here is an example of what our agent-ready looks like. Yours may look different.

    Example operation
    # Gets upcoming launches with pad visibility
    query GetUpcomingLaunchesWithPadVisibility(
      # The number of launches to return, not to exceed 10
      $limit: Int = 5
    ) {
      upcomingLaunches(limit: $limit) {
        id
        name
        pad {
          name
          weather {
            visibility
          }
        }
      }
    }

Step 2: Save the tool locally

  1. In your project, create a folder called tools.
  2. In this folder create a file with the same name as your tool. For example, GetLastFiveLaunches.graphql.
  3. Copy the from Step 1 above into this file and save it.

Step 3: Run your project as an MCP Server

To test the MCP server, we'll use the MCP Inspector.

Note: You can also use Postman to test your MCP Servers.

  1. Exit goose using CTRL + C.

  2. Run this script to kill your open port:

    lsof -t -i :4000 | xargs kill -9

  3. We are going to start with MCP enabled this time. Run this script:

    rover dev --supergraph-config supergraph.yaml --mcp mcp-config.yaml

  4. Open a new terminal in your IDE. For VSCode, you can click the + to the top right of your terminal. Run this command to start the MCP Inspector tool.

    npx @modelcontextprotocol/inspector http://127.0.0.1:8000/mcp --transport http

  5. Open the URL from your terminal output in your browser. It should look something like this:

    Example terminal output
    Starting MCP inspector...
    ⚙️ Proxy server listening on port 6277
    🔑 Session token: ...
    

    Open inspector with token pre-filled:
    http://localhost:6277/?MCP_PROXY_AUTH_TOKEN=...

  6. In the MCP Inspector, on the left sidebar, ensure that the endpoint is set to http://127.0.0.1:8000/mcp and the transport type is set to Streamable HTTP.

  7. Click the Connect button.

  8. Click List Tools. You should see your tool name.

  9. Click on the tool name to see the tool details. You may need to add a value for the that your requires. For example, if your operation requires a $limit , you may need to add a value for it.

  10. Click the Run Tool button to execute the tool.

  11. You should see the tool details and the results of the tool execution.

    A screenshot of the mcp inspector running locally with an executed tool call

Task!

Tips on writing tool descriptions

Tool descriptions help AI agents understand what the tool does and how to best use it. 's declarative nature and typed schema makes this easy. You can also:

  • Provide verbose and clear (e.g. GetFiveRecentIssues vs GetIssues))
  • Add a comment above the (prefixed by #) that says exactly what the tool does and if it requires any parameters. This will override the auto-generated description from your schema.

For more information on writing tool descriptions, check out the following resources:

Tips and troubleshooting

  • If you update the MCP configuration file, you'll need to restart the MCP server for the changes to take effect.

  • If your agent fails to run the MCP server or your would like to run it manually, use this command:

    rover dev --supergraph-config supergraph.yaml --mcp mcp-config.yaml

  • In some cases, your agent may have generated a file. To use this, add the flag --router-config router.yaml to the command.

    rover dev --supergraph-config supergraph.yaml --mcp mcp-config.yaml --router-config router.yaml

Practice

  • Can you expand on the current GetLaunches tool to include more ?
  • Instead of adding even more to the GetLaunches tool, use the Sandbox to create a new tool and save it to your tools folder.
Previous