Feature ideas

When interacting with an AI agent via a chat widget, it would be useful to expose the "current page" context to the agent. Our use-case is that we're building multiple dashboards which each embed a subset of our metrics. E.g. * DORA - for leadership * Site Reliability * AI Adoption We want to embed an AI agent into each page to enable richer interrogation of the data - but only for the metrics shown on the current page. As the agent cannot access this information, we're forced to create distinct agents for each dashboard (even if the agent prompts are virtually identical). Granted, there is benefit to tailoring agents to each dashboard - but having the choice of either approach would be very useful.

While this is motivated primarily by my company being a google cloud shop internally, I believe this is also important for simply having the best range of options as models continue to expand in capabilities.

By connecting remote MCP servers, like Notion and Atlassian, you can extend Port AI with more capabilities, tools, and data access. For example, ask, "Can you find the latest runbook for incidents in the API gateway?" which will rely on Confluence documentation.

Allow selection among models, including those that users can bring and set up their API key to use their own models.

The local python MCP supports passing CLIENT_ID and CLIENT_SECRET, but the port hosted MCP server requires the use of a browser and cookies to authenticate. We would like to use similar flow to the API server to talk to the MCP (exchange CLIENT_ID and CLIENT_SECRET for an access token and use access token in the Authorization header).

Provide a hosted MCP solution with Oauth methods to enable easier installation experience.

Support creating self-serive actions from Port MCP. In this roll out of version 0.2.15, bringing CRUD support for actions—a major step forward on our path to fully enabling Vibe Building. In this video, I am vibe building using our guides and end-to-end with the