Expand DevOps Deploy MCP Server Tools to Enable AI-Driven Operations

Tools to be added to DOD MCP Server:

Phase 1 – Read-only intelligence (highest priority)

_list_snapshot, get_snapshot_details,

_list_ldap_groups (of Teams and the members of the group)

_list_users ( in a ldap group)

- list_components, get_component

- list_component_versions, get_version_details

- list_snapshots, get_snapshot

- get_environment_inventory (installed versions per environment)

- list_resources, list_resource_pools

- list_agents, get_agent_details (status, version, last contact, pool)

- list_deployment_history with filters (application, environment, status, date range) and pagination

- get_deployment_count that works with read-only permission (currently requires Manage System)

- get_deployment_step_output for any step, not only failed steps

- list_teams, list_roles, get_permissions (read-only)

- list_processes, get_process_definition

- get_cleanup_policy, get_system_health

- list_pending_approvals, get_approval_status

Phase 2 – Guided actions (should require explicit confirmation or approval)

- restart_agent

- request_deployment / deploy_snapshot with a dry-run or validate mode

- retry_failed_deployment, cancel_deployment

- create_incident (ticketing integration)

- approve_or_reject_gate

- add_deployment_comment

Phase 3 – Autonomous operations

- subscribe_to_events (webhook or stream for deployment failed, agent offline, etc.) to avoid polling

- get_deployment_metrics (duration, failure rate, trends)

- get_failure_analysis (server-side error classification)

- rollback_to_snapshot

- set_agent_maintenance_mode

Why it's useful

1. Faster failure triage. The AI pulls the failed step, its logs, and the history in seconds. Today someone clicks through the DOD UI and reads logs by hand.

2. Answers on demand. Questions like "what version is in prod?", "which agents are offline?", or "what's blocking this deployment?" get an immediate answer, with no tickets or searching.

3. Less operator toil. At about 10,000 deployments a month, status checks and repetitive lookups add up to a lot of staff time.

4. Governance visibility. The AI can check team, role, and cleanup-policy hygiene across the environment.

5. Safe, staged adoption. Read-only comes first and carries no risk. Guided actions (restart an agent, retry a deployment) come next, with approval. Proactive monitoring comes last.

6. Less custom code. Native tools replace the workarounds you'd otherwise build and maintain, such as the credential-reuse companion service. They also give you proper scoped authentication and audit logging.

7. Proactive detection. Event subscriptions let the AI flag a failed deployment or an offline agent right away, instead of someone noticing later.

Why IBM benefits too (this helps the IDEA)

• It makes DOD more competitive as AI-ready tooling.

• Other customers with the same MCP and AI goals would use the same tools.