04 · AI and MCP
Give AI approved models, tools, context, and limits.
NopsAI separates provider credentials, model selection, agent behavior and MCP tool access into managed profiles. Pipeline authors reference approved capabilities instead of defining arbitrary access.
How it works
AI and MCP, end to end.
The flow stays explicit enough for operators to inspect, secure, automate and improve.
Register approved providers
Operators configure provider metadata and credential references without exposing secret material to pipeline authors.
Compose profiles
Agent profiles select the persona and instruction text, LLM profiles select the provider and model client, and MCP profiles select servers and allowed tools.
Resolve allowed context
Knowledge documents and hosted tools are filtered against the caller, team, scope and requested action.
Execute with evidence
Tool calls, confirmations, AI usage and resulting outputs stay connected to the conversation or pipeline run.
Platform capabilities
What teams can standardize.
These are current NopsAI platform primitives, not a future-state feature list.
Per-step model choice
Select an LLM profile at pipeline, step, task, assistant or final-output level based on the work.
Approved MCP profiles
Limit a goal to named servers and tools instead of granting arbitrary discovery across the environment.
Permission-bound assistant
Expose first-party platform operations through hosted MCP tools filtered by live AAA decisions.
Proposal and confirmation modes
Keep GitOps changes proposal-only and require explicit confirmation for side-effecting actions.
Configuration
Declared once, reviewed like code.
The example is declarative and Git-reviewable. Runtime behavior still passes through live AAA, credential, compatibility, execution and monitoring boundaries.
default_profile: devops-engineer
agent_profiles:
- id: release-manager
display_name: Release Manager
description: Coordinates release readiness and rollout.
enabled: true
instructions: |
Focus on release readiness, change risk, evidence and
rollback plans. Use configured policy and run evidence.
Never claim a change was applied. Keep decisions
traceable and operationally safe.Enterprise controls
Configuration stays inside the governance model.
Credential references
Profiles use encrypted, versioned credentials by reference; secrets are not copied into workflow definitions.
Tool allowlists
MCP profiles constrain servers and tool names, while AAA applies subject and resource policy at call time.
Bounded planning
Planner iterations, tool calls, argument sizes and confirmation requirements are limited by platform policy.
AI usage monitoring
Model, feature, run, schedule and assistant usage can be analyzed alongside cost and reliability signals.
Related capabilities
Continue through the operating model.
05 · Runtime and execution
Dispatch per-run agents to Docker or Kubernetes runners using scope, capacity, affinity, routing and runtime pools.
06 · Evidence and monitoring
Connect run history, system logs, outputs, AI usage, reliability, security, cost, alerts and recommendations.
01 · Workflow orchestration
Combine scripts, LLM goals, approvals, reusable steps and child pipelines without splitting the operating model.
Bring a real ai and mcp workflow.
We can map the configuration, authorization, runtime and evidence boundaries against your current platform model.