Why Agent Extensibility Needed a Single Home

▶ Watch (02:39)

Before Tools, makers had two separate entry points: the Add Action page inside an agent and the Library page at the environment level. Each offered a slightly different set of options. Makers weren’t sure which route to take.

The new Tools page collapses both into one place. A tool created there is visible across the entire environment, so the same REST API or flow can be attached to multiple agents without rebuilding it each time. Makers publish a tool from that page and add it to any agent in a few clicks.

Six Tool Types Available at Launch

▶ Watch (05:41)

Selecting “Create New Tool” on the Tools page surfaces six options. Prompts let makers define AI capabilities in natural language and specify input types (text, documents, images) and output format. Agent flows automate sets of business-specific tasks and come pre-wired with an agent trigger and agent response. Custom connectors wrap any REST or SOAP API not already in the 1,000-plus pre-built connector catalog. REST APIs accept a Swagger file upload directly. Computer Use gives the agent a web browser it controls via image recognition. Model Context Protocol connects to existing knowledge servers from within Copilot Studio.

“our tools are not one-size-fits-all.” — John Cartwright

Connecting to Real Systems: REST API and Prompt Builder Demo

▶ Watch (13:08)

Taylor drags and drops a Swagger file, and the tool name and description prefill automatically. She adds an API key, clicks “Next,” and the REST API tool creation from Swagger upload (13:17) publishes to the Tools page. Copilot author authentication means no re-entering credentials mid-chat.

For policy questions, she types her intent into Draft with Copilot inside Prompt Builder. A detailed prompt generates in seconds, pointing at three Dataverse columns. She switches from the default GPT-4o mini to GPT-4.5 via Bring Your Own Model, which surfaces 1,800-plus Azure AI Foundry models, then connects the prompt to her agent.

Automating Repetitive Tasks: Agent Flows and Code Interpreter

▶ Watch (19:45)

Taylor’s last tool is an agent flow for equipment orders. She adds two parameters (candidate name, start date), drops in a Freshdesk connector action, fills the ticket subject and description with those parameters, sets priority to Medium, and publishes. In the Jane Doe laptop order via Freshdesk (22:50) end-to-end test, the agent queries the HR system, Taylor confirms the two parameters, and a fourth ticket appears in Freshdesk live.

The Code Interpreter chart generation (25:24) adds one more: Taylor asks the agent to chart vacation policy data, Python runs in the backend, and a finished chart appears in seconds.

Notable Quotes

It’s the UI for AI. Yogi Naik · ▶ 1:09

Tools is one-stop shop for best-in-class features Yogi Naik · ▶ 4:25

our tools are not one-size-fits-all. John Cartwright · ▶ 3:35

Key Takeaways

  • Tools in Copilot Studio (public preview) unifies agent extension into one page at the environment level, not just per-agent.
  • Six tool types cover prompts, agent flows, REST APIs, custom connectors, Computer Use, and Model Context Protocol.
  • Bring Your Own Model lets makers swap in any of 1,800-plus Azure AI Foundry models inside a prompt tool.