blissful-infra mcp

The MCP server exposes blissful-infra’s tenant model and Kubernetes golden path as tools for Claude. Use it to let Claude scaffold services, run deploys, manage canaries and query cluster state.

blissful-infra mcp

The server listens on stdin/stdout as JSON-RPC 2.0 and emits everything to stderr (so logs don’t pollute the stream). Point Claude to it via your client’s MCP configuration.

Installation

From npm (2.0 and later)

Once blissful-infra 2.0 ships to npm, install globally:

npm install -g @blissful-infra/cli
blissful-infra mcp

Or as a local dev dependency in your project:

npm install --save-dev @blissful-infra/cli
npx blissful-infra mcp

From GitHub (development / unreleased)

For the bleeding-edge version, clone the repo and run from source:

git clone https://github.com/cavanpage/blissful-infra.git
cd blissful-infra

# Install dependencies
npm install

# Build the CLI
npm run build

# Run the MCP server
node packages/cli/dist/index.js mcp

You can also use npm link to simulate a global install without publishing to npm:

cd blissful-infra/packages/cli
npm link

# Now 'blissful-infra' is available globally as if installed from npm
blissful-infra mcp

Configuring Claude to use the MCP server

The exact configuration depends on your Claude client. Here are the common patterns:

Claude Code (desktop or VS Code)

Add to your Claude Code settings file (location depends on your client):

{
  "mcpServers": {
    "blissful-infra": {
      "command": "node",
      "args": ["/path/to/packages/cli/dist/index.js", "mcp"]
    }
  }
}

Or if you used npm link:

{
  "mcpServers": {
    "blissful-infra": {
      "command": "blissful-infra",
      "args": ["mcp"]
    }
  }
}

Claude.ai (web)

If using Claude.ai with local MCP forwarding (e.g., via a tunnel or local proxy):

{
  "mcpServers": {
    "blissful-infra": {
      "command": "sh",
      "args": ["-c", "cd /path/to/blissful-infra && npm run build && node packages/cli/dist/index.js mcp"]
    }
  }
}

What tools does Claude get

The MCP server exposes 21 tools covering:

Category	Tools
Discovery	`list_tenants`, `get_tenant`, `list_projects`, `get_project`, `list_services`, `get_service`
Lifecycle	`create_tenant`, `delete_tenant`, `create_project`, `delete_project`, `add_service`, `remove_service`
Kubernetes golden path	`cluster_up`, `cluster_down`, `cluster_status`, `deploy_service`, `canary_status`, `canary_promote`, `canary_abort`
Observability	`get_logs`, `get_metrics`, `list_ports`

Each tool takes tenant/project/service coordinates and returns structured data. Long operations (cluster up, deploy) return a job ID polled via get_job.

Example: deploy a service via Claude

With the MCP server running, you can ask Claude:

“Deploy the orders service to our prod cluster with a canary rollout.”

Claude will:

Call list_services to find the orders service
Call deploy_service with the service coordinates
Poll get_job until the deploy completes
Call canary_status to show you the rollout state
Wait for your signal to promote