I had too many AI tools and one web portal trying to connect them.
The first integrations included Slack, Gmail, Notion, and Linear. The portal handled the connection setup and made selected actions available from a Claude conversation.
The goal was not to replace each service's interface. It was to make ordinary cross-tool tasks possible without opening every application separately. Find a message, draft a reply, look up a document, or create a work item from the same conversational surface.
I was also testing delegation. A connection should not automatically grant another person every action available to the account owner. The portal needed a smaller permission set that could allow reading or drafting while withholding sending, deletion, or administrative access.
Setup was part of the product. Model Context Protocol servers could expose the tools, but a person still needed to authorize each underlying service and understand what the assistant could do afterward.
The interface was early. I asked another designer for feedback on where the connection process became unclear and which account states needed better explanation.
The private source contains tokens, internal endpoints, account identities, and implementation detail. None of those values are included here. This article does not describe the portal as a finished or publicly released service.
The public-safe event is the working prototype. By March 8, I could use one web setup to connect multiple services and make their tools available inside Claude.
I continued testing additional sources, including professional messages, while keeping each connection separate enough to disable or change later.
Each service also displayed its own connection status.
That Sunday, the portal was useful enough to demonstrate and unfinished enough that the next work was still basic: improve onboarding, clarify permissions, and make it obvious which connected account would answer a request.