Development to Production Workflows

This guide outlines how to work with your Apollo MCP Server in development, staging, and production environments. For each environment, your workflow differs in how you define and manage your tools, deploy your server, and configure additional settings.

• Local development: Define operations as local files, run your MCP server locally with rover dev, and test.

• Dev and staging environments: Publish schema to graph variants, manage tools using operation collections, deploy and test.

• Production environments: Choose your operation source (files or persisted queries), configure authentication, deploy, test, and monitor.

Local development

1. Define tools using local operation files.

   YAML

   Example MCP config file

   copy

   1operations:
   2  source: local
   3paths:
   4  - ./operations/local.graphql

2. Run your MCP server locally using rover dev, alongside your GraphQL API.

   Bash

   copy

   1rover dev --supergraph-config supergraph.yaml --mcp .apollo/mcp.local.yaml

   You can also run the MCP server using Docker or the binary.

3. Test your tools using MCP Inspector or connect the server to an AI agent.

Dev and staging environments

1. Create separate graph variants for dev and staging.

2. Publish your schema to each variant.

3. Define tools using operation collections.

   Create and organize operations per graph variant. Then, configure your MCP server to use the collection.

   YAML

   Example MCP config file using operation collections

   copy

   1operations:
   2  source: collection
   3  id: default

4. Deploy an MCP server pointing to each variant.

5. Connect your AI agent to each environment and validate behavior.

Production environments

1. Define your tools using either local operation files or persisted queries.

   Choosing the right source for your tools

   Using local operation files

   • Store operations in your git repository
   • Operations are version-controlled and reviewed in PRs before deployment
   • Use CI/CD to validate and deploy
   • See Define Tools > From operation files

   YAML

   Example MCP config file

   copy

   1operations:
   2  source: local
   3paths:
   4  - ./operations/local.graphql

   
   
   Using persisted queries

   • Publish operations to a persisted query list
   • Configure safelisting with persisted queries, so that the router enforces that only registered operations can execute
   • Use CI/CD to publish operations to the persisted query list
   • See Define Tools > From GraphOS-managed persisted queries

   YAML

   Example MCP config file

   copy

   1operations:
   2  source: uplink

   You will also need to set your graph credentials APOLLO_GRAPH_REF and APOLLO_KEY as environment variables.

2. Configure authentication in your MCP server.

3. Disable introspection tools.

4. Deploy the MCP Server using container-based deployment to your infrastructure.

5. Monitor your MCP server using:

   • GraphOS Studio: View operation metrics and usage patterns in Insights

   • OpenTelemetry: Instrument for traces, metrics, and logs by configuring telemetry