> For the complete documentation index, see llms.txt.
Skip to main content

Check out Port for yourself ➜ 

Agent management


Agent management is about governing every AI agent operating in your organization - what data they can access, what actions they can take, and who can use them. That includes code-first agents you build with SDKs in your own repos or CI (for example OpenAI Agents SDK, Anthropic Claude Agent SDK, or Claude Code), cloud-managed agents hosted on vendor control planes (for example AWS Bedrock / AgentCore, Azure AI Foundry, or Cursor Cloud Agents), and Port-native agents configured inside Port.

What roles do AI agents play in Port?​

Agents interact with your software factory in three complementary ways. Understanding these patterns helps you decide when to let an existing agent consume Port, when the platform should trigger agents on its own, and when to provision a new agent.

AI agents as platform consumers​

The most intuitive role is the agent as a user of the platform. For example, an engineer asks a coding agent to add an endpoint to a payments service. Before writing code, the agent pulls the service's owner, dependencies, and standards from Port, then spins up a preview environment and runs tests through self-service workflows.

Claude Code agent reads context from Port and runs platform actions such as creating a preview environment and running tests

Here the agent works side by side with the developer. It reads from Context Lake and runs governed actions, the same way a human would use the platform. Use this pattern when developers already work in tools like Claude Code, Cursor, or GitHub Copilot. The Port MCP server exposes catalog context and workflows so those agents operate with the same source of truth and guardrails as humans.

AI agents as internal platform components​

The next level treats the agent as part of the platform engine, not only as a user of it. For example, when a vulnerability is opened, Port can take a remediation agent from the agent registry (or invoke an agent for that case without using a registered one) and run it inside a workflow to remediate across affected services.

Agent registry remediation agent triggered by a nightly scan, opening pull requests across many services

Nobody has to request an agent for each case. The platform triggers the agent from an event, and owning teams wake up to outcomes such as open PRs. Use this pattern for autonomous work such as security remediation, self-healing incidents, or other event-driven loops where the agent is a pillar of the platform itself.

AI as reusable building blocks​

The third role is AI as a reusable building block. For example, an engineer needs an agent. They pick the model, tools, network landscape, and use case, and the platform hands back an agent that already sits on the golden path, with identity, approved tool access, and the right scope.

Engineer requests an agent via a self-service workflow and receives a registered, governed agent with identity and tool access

"Agent" here also covers related AI artifacts such as LLMs, MCP servers, and skills. Teams often struggle to enforce the right tools and access, apply guardrails the way they would for any other service or resource, and hand over an agent once it is built. Self-service workflows such as Create an agent, Onboard an LLM, Add MCP server, and Publish a skill turn that lifecycle into a governed process for creating, registering, and sharing AI artifacts across the organization.

Govern agents, skills, and MCP at scale​

As engineering teams adopt AI agents, the number of agents, skills, prompts, and MCP servers grows quickly. Without a shared layer to discover and govern them, each team operates in isolation, and platform teams lose the ability to enforce standards or answer basic operational questions.

Without shared governance, agents, skills, prompts, and MCP servers end up scattered across teams, making it hard to see what exists, control access, or avoid duplication. Each team maintains separate configurations and there is no single place to track ownership, security, or approved tools.

How Port helps​

Port gives you one place to discover and govern agents, skills, prompts, and MCP across your organization:

You need to…Port capability
Route and govern model traffic at scale, then connect gateway data to teams and servicesPort and AI gateways (AgentOps)
Discover agents on external platforms, code-first (OpenAI Agents SDK, Claude Code, GitHub Actions) or cloud-managed (Bedrock / AgentCore, Azure AI Foundry, Cursor Cloud Agents)External agents
Maintain a registry of agents, skills, and MCP servers with ownership and status in Context LakeAgentic registry (AgentOps)
Publish skills and prompts, and control which external MCP tools teams may useSkills, prompts, and MCP connectors
Give each agent a curated, scoped set of tools from Port and MCP connectorsVirtual MCP servers
Build and run agents natively in PortPort custom agents
Scope what agents can read from Context LakeContext Lake data access
Connect IDEs and external tools to your Context LakePort MCP server

Port brings agents, skills, prompts, and MCP under one governance layer so teams can move fast without losing control.

What's in this section​

External agents​

Connect and govern agents hosted outside Port, whether you build them with SDKs in code or run them on cloud-managed platforms. Discover agents in Context Lake, trigger sessions from workflows, and track deployments through an agent registry.

Explore external agents β†’

Port MCP server​

Connect IDEs, AI tools, and agents to Port using natural language. The MCP server exposes your Context Lake and workflows as callable tools, so agents like Claude, Cursor, and GitHub Copilot can query the catalog and take governed actions from wherever developers already work.

Learn more about the Port MCP server β†’

Port custom agents​

Customize and orchestrate complicated workflows inside Port. Build intelligent agents that can be used as part of automations and engineering workflows.

Use Port custom agents to:

  • Automate incident response workflows
  • Create intelligent PR review processes
  • Build custom task management assistants
  • Generate automated deployment reports
  • Orchestrate multi-step engineering processes

Explore Port custom agents β†’

Skills​

Publish reusable instructions that Port AI and MCP clients load at runtime, and measure which skills teams adopt in Cursor and Claude.

Learn about skills β†’

MCP connectors​

Connect external MCP servers to Port and choose which of their tools developers and agents can use, with the same permissions and audit trail as the rest of Port.

Explore MCP connectors β†’

Virtual MCP servers​

Give each agent an MCP endpoint that exposes only the tools it needs, renamed for the job and with arguments fixed ahead of time.

Learn about virtual MCP servers β†’

Govern agentic assets across your organization

To build a registry of your agents, skills, and MCP servers, certify them, and track their ROI, see the AgentOps solution.

FAQ​