holmes-mcp-integrations

MCP integrations for HolmesGPT

Repository Structure

holmes-mcp-integrations/
├── build-all-mcp-servers.sh      # Build script for all MCP servers
├── mcp_base_image/               # Base Docker image for MCP servers
└── servers/                      # MCP server implementations
    ├── aws/                      # AWS API MCP Server
    ├── aws-multi-account/        # AWS Multi-Account MCP Server
    ├── azure/                    # Azure CLI MCP Server
    ├── confluence/               # Confluence MCP Server (external image)
    ├── gcp/
    │   ├── gcloud/               # GCP gcloud CLI MCP Server
    │   ├── observability/        # GCP Observability MCP Server
    │   └── storage/              # GCP Storage MCP Server
    ├── github/                   # GitHub MCP Server
    ├── kubernetes-remediation/   # Kubernetes Remediation MCP Server
    ├── mariadb/
    │   └── mcp-minimal/          # MariaDB MCP Server (minimal)
    └── sentry/                   # Sentry MCP Server

Building MCP Servers

Configuration

Each MCP server directory contains an auto-build-config.yaml file that defines the image name and version:

image: azure-cli-mcp
version: "1.0.2"

Build All Servers

Use the build-all-mcp-servers.sh script to build and push all MCP server images:

# Build all servers (skips images that already exist)
./build-all-mcp-servers.sh

# Dry run - list servers and check which images exist (no build)
./build-all-mcp-servers.sh --dry-run

# Force rebuild all servers (even if they exist)
./build-all-mcp-servers.sh --force

# Show help
./build-all-mcp-servers.sh --help

The script will:

• Build the base image first (mcp_base_image/supergateway_base)
• Auto-discover all servers with auto-build-config.yaml
• Check if each image already exists in the registry
• Skip existing images (unless --force is used)
• Build multi-platform images (linux/amd64, linux/arm64)
• Push to the configured registry

Base Image

The base image (supergateway_base) is built first and tagged with both version and :latest. All MCP servers depend on this base image. Its config is in mcp_base_image/auto-build-config.yaml.

Build Single Server

Each server directory also has its own build-push.sh script:

cd servers/azure
./build-push.sh

Bumping Versions

To release a new version of an MCP server:

1. Update the version in the server's auto-build-config.yaml:

   image: azure-cli-mcp
   version: "1.0.3"  # bumped from 1.0.2

2. Run the build:

   # Build just that server
   cd servers/azure && ./build-push.sh
   
   # Or build all (will only build the changed one)
   ./build-all-mcp-servers.sh

Registry

All images are pushed to:

us-central1-docker.pkg.dev/genuine-flight-317411/mcp/

Adding a New MCP Server

1. Create a new directory under servers/:

   mkdir servers/my-new-mcp

2. Add a Dockerfile:

   FROM us-central1-docker.pkg.dev/genuine-flight-317411/mcp/supergateway_base:latest
   # ... your server setup

3. Add auto-build-config.yaml:

   image: my-new-mcp
   version: "1.0.0"

4. Add build-push.sh:

   #!/bin/bash
   set -e
   SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
   REGISTRY="us-central1-docker.pkg.dev/genuine-flight-317411/mcp"
   
   image=$(grep '^image:' "$SCRIPT_DIR/auto-build-config.yaml" | sed 's/image: *//')
   version=$(grep '^version:' "$SCRIPT_DIR/auto-build-config.yaml" | sed 's/version: *//' | tr -d '"')
   
   docker buildx build --pull --no-cache --build-arg BUILDKIT_INLINE_CACHE=1 --platform linux/arm64,linux/amd64 --tag "$REGISTRY/$image:$version" --push .

5. Make it executable:

   chmod +x servers/my-new-mcp/build-push.sh

The new server will be automatically discovered by build-all-mcp-servers.sh.