← Back to all sparks
P

Port

INFRA · APIS
Velocity3.8

Internal developer portal for platform engineering, service catalogs, and scorecards.

Port AI graduated from answering questions about the catalog to changing it, with approval.

agentic sdlcinternal developer portalmcpworkflow automationextensibilitybyo llm
Current state
Port is an internal developer portal whose center of gravity has moved to Port AI. Over ten months it went from an open-beta chat assistant to an agent that plans and executes real platform changes — creating blueprints, upserting entities, triggering workflows — behind a plan the user approves first. The surrounding platform work is almost entirely extensibility: a public plugins repo, Custom Widgets, Workflows in open beta, and MCP connectors that pull external tools into the same chat.
Where it's heading
The arc is assistant to gateway to executor. March made Port AI a front end for other vendors' tools via external MCP servers; June put a visual automation builder underneath it; July let the AI drive both, with Plan and Build modes and per-step traces. Each release widens what the agent may touch while keeping a human approval gate in front of it, and the extensibility work — plugins, widgets, bring-your-own LLM endpoint — exists so customers can extend the surface the agent operates on.
Prediction
Expect the approval gate itself to become the product surface: policy governing which plans may execute unattended, tied to the Dynamic RBAC and permissions simulator already shipped. The step-by-step plans and per-step traces in the July release are the groundwork for granting scoped autonomy.

Recent moves

  1. 6d ago

    Port AI Builder plans and executes catalog changes with approval

    ⚡ SPARK

    This is the release the previous ten months were building toward: Port AI moves from describing the catalog to changing it. Plan mode proposes a step-by-step change set, Build executes it incrementally with traces, and the approval gate is what keeps the whole thing inside the platform team's control.

  2. 1mo ago

    Workflows hits open beta; AI agents connect external MCP servers

    Workflows reaching open beta gives the automation layer a visual, node-based builder instead of scattered action configs — the substrate July's AI Builder would later drive. External MCP servers inside AI agents extend March's gateway work from the side chat into the agents themselves, and Private Pages let individuals build catalog views without publishing them.

  3. 2mo ago

    Public plugins repo opens Port to community contributions

    The port-plugins repo makes Port's extension points public and forkable, which the notes themselves frame as an early step rather than a finished ecosystem. Claude Code and Copilot integrations pull AI coding activity into the catalog, consistent with Port positioning itself as the place engineering activity gets measured.

  4. 3mo ago

    Custom Widgets run your own UI inside Port; BYO OpenAI endpoint

    Custom Widgets let customers build frontend components in their own stack and run them inside Port dashboards and entity pages, while support for any OpenAI-compatible endpoint loosens the model dependency in Port AI. Both are extensibility moves that widen the surface the later agentic releases would operate on.

  5. 4mo ago

    MCP Connectors turn Port AI into a gateway for external tools

    ⚡ SPARK

    MCP Connectors reposition Port AI from a catalog assistant into a single front end for the whole tool stack. It is the first release where Port's AI value comes from what it can reach rather than what it knows about your catalog.

  6. 5mo ago

    Run Claude in your own Azure; RBAC gets a UI and simulator

    Running Claude models inside a customer's own Azure environment addresses the data-residency objection that stalls AI adoption in regulated orgs. The Dynamic RBAC UI, permissions simulator, and secrets audit logs are the governance groundwork the later agentic releases lean on.