July’s not even over yet and we’ve already released four significant new agent-optimized toolkits to Arcade.dev, plus meaningful upgrades to Gmail, Excel, Outlook, and Linear.
Here’s what your agents can do now that they couldn’t a month ago.
Power BI: Intuitive data and analytics
With Microsoft Power BI agents can confirm their signed-in identity, discover workspaces, datasets, and reports, inspect a semantic model’s schema, and run or validate a Data Analysis Expressions (DAX) query against it. That means an agent can pull real numbers out of a Power BI dataset without anyone opening the portal. It can also check a dataset’s refresh history, trigger a refresh, and confirm a report reflects current data before reading from it.
This one came straight from a customer. Their day-one priority was getting Power BI into Claude fast and giving their data teams access to semantic models in Cowork. Microsoft’s remote MCP server for Power BI is still in beta, though, and it made them supply a model’s globally unique identifier (GUID) just to ask something as simple as “show me last quarter’s revenue.” We took their team’s requirements, ran them through our tools pipeline, and shipped an optimized Power BI toolkit that just works.
The first time someone on their team asked a plain-language question, Claude came back with the right answer in seconds. This is a great example of what we mean when we say well-built tools can make AI “actually” useful for businesses.
Postman: Build, test, document, and manage APIs
Postman gives agents a way to manage API work end to end. They can browse workspaces, save and fork collections, export a collection as OpenAPI, and build out folders and requests inside one. They can also stand up mock servers for testing, then create, run, and read the results of monitors to check collection health on a schedule.
Fireflies: AI transcription and recording
With Fireflies agents can find a recorded meeting by title, date, attendees, or topic and read its verbatim transcript or AI summary, then ask grounded questions and follow up without reading the whole thing. They can also gather action items across recent meetings into one list, pull per-speaker talk-time and sentiment, clip soundbites, and send the notetaker into a live call.
Insightly: Flexible CRM
Insightly brings your CRM into the conversation. Agents can find, create, and update contacts and organizations, capture and convert qualified leads into a contact, organization, and opportunity in one call, and manage opportunities by advancing stages, recording won or lost, and rolling a pipeline up by value. They can also log notes and create tasks against the deal, contact, or project they belong to.
Other upgrades worth noting
Excel workbooks now come formatted by default. When an agent creates a workbook, it builds a properly typed, native table with number formats and autofit columns, so the output looks polished without a separate formatting pass.
Gmail picked up a batch of improvements. Agents can trash or restore an entire thread in one call, forward a message with its original attachments and a note, and search across multiple sender, subject, or body terms at once. One upgrade matters for security in particular. Gmail now strips hidden text from an email’s HTML (white-on-white or display:none content) before an agent reads it, so a sender can’t smuggle invisible prompt-injection instructions into what looks like a normal message.
Outlook can now draft a reply or reply-all without sending it, in your own mailbox or a shared one, so a human reviews the message before it goes out. And Linear can fetch a project’s posted status updates, so an agent catches up on progress without leaving the conversation.
Why “agent-optimized” matters
Every one of these tools is hand-built for the way agents actually work, not wrapped around a raw API. They run on Arcade’s runtime with delegated user authorization, so an agent acts scoped to exactly what that user is permitted to do, evaluated per action. A support agent reads the tickets that user can read. A Power BI agent queries the datasets that user can query. Nothing more.
Try them at arcade.dev and tell us what you build.
