Extend team agents beyond Code: connect Chrome, Excel, PowerPoint into one coordinated team.

Resolved 💬 4 comments Opened Feb 25, 2026 by andrewleech Closed Apr 21, 2026

Preflight Checklist

  • [x] I have searched existing requests and this feature hasn't been requested yet
  • [x] This is a single feature request (not multiple features)

Problem Statement

The new team agent feature in Claude Code is really exciting. Being able to spin up sub-agents that communicate bidirectionally opens up a lot. But right now that communication is scoped to Code-to-Code. Meanwhile Anthropic is shipping Claude into Chrome, Excel, and other products that each have their own specialized tooling. These products are siloed: they can't talk to each other or share context.

The problem is that real workflows don't live in one product. I want to research something on the web, pull data into a spreadsheet, run some analysis, generate code to transform it, and build a presentation. Right now each of those steps is a completely separate conversation with no shared state.

Proposed Solution

Generalize the multi-agent communication protocol so Claude Code can act as an orchestrator that other Claude products can connect to. Think of it like Code advertising itself as a hub, and Claude in Chrome, Claude in Excel, Claude in PowerPoint, etc. registering as specialized agents in a team.

The key capabilities I'm imagining:

  • Data flow between products. Code uses filesystem tools to assemble or transform data, hands it to Excel for charting, which feeds into PowerPoint for a presentation. Structured handoffs, not copy-paste.
  • Tool assistance across products. Excel or Chrome could ask Code to run a script, hit an API, or do filesystem work on their behalf. Code has the most powerful tooling, so letting other products leverage that conversationally feels like a natural extension.
  • Shared context. Agents in the team can see what others are working on and coordinate, the same way sub-agents can today within Code.

The motivating example: I ask Claude Code to orchestrate a research report. Chrome goes and gathers information, Code processes and structures the data, Excel builds the charts, PowerPoint assembles the deck. One prompt, multiple specialized agents, coherent output.

Alternative Solutions

Today you'd do each step manually in separate products, copy-pasting between them. Or you'd try to do everything in Code alone, but then you lose the specialized UI and tooling that Excel/Chrome/etc. bring. MCP servers help with some of this but they're tool-level integrations, not agent-level coordination.

Priority

High - Significant impact on productivity

Feature Category

Other

Use Case Example

I tell Claude Code: "Research the current state of RISC-V adoption in embedded systems and put together a presentation with supporting data. Finance wants the raw data in Excel too."

  1. Code delegates research to Claude in Chrome, which browses vendor announcements, conference talks, and market reports, summarizing findings back to the orchestrator. Chrome also fills out a request form on a market data portal to get access to a gated dataset.
  2. Code takes those findings and writes a Python script to pull and clean publicly available data (GitHub repo counts, SDK download trends, vendor ecosystem stats) into structured CSVs.
  3. Code hands the CSVs to Claude in Excel, which builds a formatted workbook with pivot tables, named ranges, and conditional formatting that the finance team can open and work with natively.
  4. Code sends the research summary and key figures to Claude in PowerPoint, which assembles a deck with an executive summary, key findings, and supporting visuals.

I get a coherent deliverable from one prompt, where each product did what it's best at: Chrome handled web interaction including form submission, Code did the heavy data processing, Excel produced a handoff-ready workbook for another team, and PowerPoint built the presentation.

Additional Context

The sub-agent protocol already handles bidirectional messaging and coordination between agents. This proposal is essentially saying "make the discovery and connection part product-aware so the same protocol can bridge across Anthropic's product suite." The hard part is probably less the messaging and more the shared state and data handoff semantics, but even a basic version where Code can push files to Excel or request Chrome to fetch something would be hugely valuable.

View original on GitHub ↗

This issue has 4 comments on GitHub. Read the full discussion on GitHub ↗