Ecosystem Bootstrap

Ecosystem Bootstrap Workflow

Self-deploying bootstrap that unfolds agent-utilities and the full agent ecosystem on a fresh machine or server. Walks through deployment profile selection, prerequisite installation, core infrastructure, service containers, MCP server containers, DNS routing, and Knowledge Graph registration.

Steps

Step 0: select-deployment-profile [skill: user-interaction]

Canonical profiles — agent-utilities day 0 uses three current tiers: tiny (all-local, zero-infra), single-node-prod (one durable host), and enterprise (multi-host Kubernetes). Do not translate legacy profile names. The *-mcp connector set per profile is the single source of truth in agent-utilities/deploy/mcp-fleet.registry.yml.

Present the deployment profile questionnaire to the user. Ask:

1. Deployment profile: tiny, single-node-prod, or enterprise
2. Target host: localhost, remote SSH host (via tunnel-manager), or multi-node cluster
3. DNS profile reference: AgentConfig connection/policy reference for service routing
4. GPU availability: whether to deploy GPU-accelerated services (data-science-mcp, CUDA models)
5. External services: which cloud services are already configured (Jira, ServiceNow, GitHub, etc.)
6. Storage paths: data volume mount points for persistent storage Expected: deployment_profile, target_host, dns_domain, gpu_available, external_services, storage_config

Step 1: systems-manager-mcp

Verify and install system prerequisites on the target host:

• Docker Engine + Docker Compose v2
• Python 3.11+ and pip/uv
• Git for repository cloning
• SSH server (if remote deployment via tunnel-manager)
• Network utilities (curl, wget, jq) Use systems_manager_execute to run installation commands. Verify each dependency with version checks. Expected: docker_version, python_version, git_version, system_ready Depends On: Step 0

Step 2: container-manager-mcp

Deploy Tier 0: Core Infrastructure containers directly via container-manager-mcp (Portainer isn't available yet):

• Portainer CE — container orchestration UI (port 9443)
• Technitium DNS — authoritative DNS server (port 5380/53)
• Traefik (optional) — reverse proxy for service routing Use container_manager_docker with action='docker_create_container' for each. Create the shared Docker network agent-net first via action='docker_create_network'. Expected: portainer_container_id, dns_container_id, agent_network_id Depends On: Step 1

Step 3: verify-portainer [skill: portainer-mcp]

Verify Portainer is accessible and complete initial setup. Use portainer_auth to authenticate, then portainer_environment with action='get_endpoints' to confirm the local Docker endpoint is registered. All subsequent container deployments will use Portainer stacks. Expected: portainer_auth_token, endpoint_id Depends On: Step 2

Step 4: configure-service-dns [skill: technitium-dns-mcp]

Configure DNS records through the AgentConfig-resolved DNS connection. Resolve the zone from that profile and use add_record once per enabled service. Pass opaque service and address references to the provider; do not retain resolved names or addresses in workflow output. Expected: dns_records, record_count Depends On: Step 2

Step 5: deploy-platform-services [skill: portainer-mcp]

Deploy Tier 1: Platform Services through the orchestrator selected by the canonical deployment profile. Use the profile's enabled component set as the source of truth; do not carry a separate environment-specific service list. Resolve image, storage, identity, network, and TLS settings from each component profile. Expected: deployed_stacks, stack_ids, service_urls Depends On: Step 3, Step 4

Step 6: deploy-mcp-services [skill: portainer-mcp]

Deploy the MCP connector set selected by the canonical fleet registry and deployment profile. Each enabled connector receives a health check and only named AgentConfig connection references; the workflow does not enumerate a local fleet. Use the orchestrator adapter selected by the profile. Expected: mcp_containers, mcp_endpoints, health_status Depends On: Step 5

Step 7: update-mcp-dns [skill: technitium-dns-mcp]

Update DNS records with the actual container IPs for all deployed services and MCP servers. Use add_record for each MCP server endpoint. Expected: mcp_dns_records, total_records Depends On: Step 6

Step 8: graph-os

Register the complete deployment topology in the Knowledge Graph. Use graph_write to create nodes:

• DeploymentManifest — the overall bootstrap record with profile, timestamp, host
• DeploymentPhase nodes for each tier (Tier 0, 1, 2)
• MCPServerDeployment nodes for each MCP server with endpoint URLs
• PlatformService nodes for each deployed service
• Create edges: deployedOn, belongsToStack, resolvesDNSFor, dependsOn, bootstrappedBy Expected: manifest_node_id, topology_nodes, relationship_count Depends On: Step 6, Step 7

Step 9: present-bootstrap-summary [skill: user-interaction]

Present the deployment summary dashboard to the user:

• Total services deployed (by tier)
• DNS routing table
• MCP server health status
• Knowledge Graph node count
• Service access URLs
• Next steps: run full-infrastructure-discovery to populate detailed container metrics Expected: deployment_report, user_acknowledgment Depends On: Step 8

Execution

Run this workflow as a dependency-ordered DAG. Steps with no unmet depends_on run in parallel; dependents run after their prerequisites complete.

• Run first (in parallel): Step 0 — select-deployment-profile; Step 1 — systems-manager-mcp; Step 2 — container-manager-mcp; Step 3 — verify-portainer; Step 4 — configure-service-dns; Step 5 — deploy-platform-services; Step 6 — deploy-mcp-services; Step 7 — update-mcp-dns; Step 8 — graph-os; Step 9 — present-bootstrap-summary

Execution: If graph-os is reachable, offload the whole DAG via graph_orchestrate action=execute_workflow (or the kg-delegate skill) for true parallel/swarm execution. Otherwise execute the steps natively in dependency order: run steps with no unmet depends_on in parallel, then their dependents.