Sign in Subscribe

Supercharge Copilot with Custom Actions and Microsoft Copilot Studio

Jamie McAllister

• 4 min read

Copilots are clever little helpers that can juggle most jobs you throw at them. Their out‑of‑the‑box tricks are handy, but they turn into absolute rock stars once you wire them up to outside data and tools. In Microsoft Copilot Studio, that party piece is called a custom action.

Plenty of power users and devs still haven’t twigged just how far these actions can take them. Copilot Studio lets you spin up new copilots, beef up the Microsoft ones you already use, and bolt on actions that fetch data, kick off workflows or answer those awkward questions the default brain simply can’t handle.

How a copilot uses custom actions

  • Automatic use via generative orchestration. Flick this switch and the copilot susses out the user’s request, then picks the best action, topic or knowledge source on its own. A tidy description of each action helps it decide when to pull the trigger and, if needed, prompts it to ask for extra details. Sorted.
  • Explicit calls inside a topic. Prefer a more scripted chat? Pop an action at a certain step and Bob’s your uncle.

Custom actions save everyone from building yet another bespoke connector. The platform already gives you standard plugs into the outside world, so you can get on with life and maybe squeeze in a cuppa. 🙂

The seven action types

1. Prebuilt connector actions

Ready‑made connectors hook straight into services like SharePoint, Outlook and Dataverse.

  • Use case: Bread‑and‑butter tasks in everyday business apps.
  • How to add: In Copilot Studio go to Actions > Add an action > Connector, then choose the connector. You may need to authorise the connection.
  • Tech note: Some connectors, especially those born from AI plugins or Dataverse APIs, want proper authentication.

2. Custom connector actions

When the API you fancy isn’t covered, roll your own connector.

  • Use case: Line‑of‑business systems, custom databases or specialist services.
  • Steps to add:
    1. Build and test the connector in Power Apps or Power Automate, usually from an OpenAPI (YAML) file.
    2. Work in an environment with Dynamics 365 apps.
    3. Configure OAuth 2.0 with Microsoft Entra ID and tick Enable on‑behalf‑of login.
    4. In Copilot Studio choose Actions > Add an action > New custom connector, then publish.
    5. An admin enables it for users (can take up to 12 hours, so grab a biscuit while you wait).
    6. Optionally submit it for certification if you want to share outside your tenant.
  • Tech note: You’ll need to know the API, keep auth tight and mind the compliance cops.

3. Power Automate cloud flow actions

Cloud flows let you string together multi‑step processes across several systems.

  • Use case: Approvals, document generation, cross‑platform notifications, data wrangling.
  • How to add inside a topic:
    1. In the authoring canvas hit the plus icon.
    2. Pick Call an action.
    3. Choose your flow.
    4. Map topic variables to the flow’s inputs.
    5. Use its outputs later on.
  • Tech note: Keep flows lean or the chat will plod. Test both the flow and the topic. Flows are handy for chopping up gnarly JSON that comes back from Bot Framework skills.

4. AI Builder prompt actions

AI Builder runs generative AI tasks based on plain‑English instructions.

  • Use case: Summaries, information extraction, classification, sentiment‑savvy replies.
  • Integration options:
    • Prompt action (Copilot extension) via the wizard.
    • Topic node using Call an action > Create a prompt.
    • AI plugin so the copilot decides when to run it.
  • Build steps: Write clear instructions, add grounding data if needed, set inputs and outputs, then test.
  • Tech note: Needs a Dataverse database. Prompts eat AI Builder credits. GPT‑4 licensing differs from GPT‑3.5, so keep finance happy.

5. Bot Framework skill actions

Plug in an existing Bot Framework skill when you need heavyweight pro‑code logic.

  • Use case: Complex bots, hefty calculations or systems that demand fine‑grained handling.
  • Prerequisites: A standalone Copilot Studio subscription (not the Teams one) and a deployed skill.
  • How to add: In a topic, select Call an action and pick the skill, then map inputs and outputs.
  • Tech note: Skills return simple types or JSON strings. If you need to pick through JSON, parse it in Power Automate first.

6. REST API connection

For straightforward REST endpoints, skip the full connector dance.

  • Use case: Quick calls to a simple API.
  • How to add: Choose Add an action and select the REST option.
  • Tech note: Know your HTTP verbs, endpoints and auth scheme.

7. Model Context Protocol (MCP) actions

MCP is an open standard that points AI models at the right data and tools, cutting down faffy integration.

  • Use case: Pulling in tools or data from an MCP server, especially when you’ve lots of sources across different vendors.
  • How it works: An MCP client inside Copilot Studio rings an MCP server. Each tool the server exposes appears as an action in Copilot Studio and updates automatically.
  • Steps to integrate:
    1. Spin up an MCP server with the SDK.
    2. Create a custom MCP connector in Power Apps or Power Automate from an OpenAPI file, tagging it Agentic and McpSse.
    3. In Copilot Studio add the connector via Actions > Add an action > Connector and authorise it.
    4. Optionally publish it for cross‑tenant use.
  • Tech note: Generative orchestration must be on. Copilot Studio currently supports Server‑Sent Events transport. Needs Dataverse. MCP also supports bidirectional magic called Sampling.

Choosing the right approach

  • Prebuilt vs custom connectors
    • Prebuilt: Fast to set up but limited to what’s already on the shelf.
    • Custom: Works with any API but requires more elbow grease and upkeep.
  • Power Automate flows
    • Best for end‑to‑end processes and data transforms.
    • Watch for lag if the flow gets too chonky.
  • AI Builder prompts
    • Gold for unstructured text.
    • Quality hinges on sound prompt design and decent input.
  • Bot Framework skills
    • Give maximum coding freedom.
    • Need dev chops and extra parsing for complex data.
  • MCP actions
    • One protocol, many tools, enterprise security baked in.
    • Requires an MCP server and connector.
  • REST API connections
    • Perfect for quick one‑offs.
    • Less reuse and governance than connectors or MCP.

Final thoughts

So there you have it, mate. Custom actions turn a decent copilot into a Swiss Army knife. Whether you grab a prebuilt connector, fire off a cloud flow, craft an AI Builder prompt, unleash a Bot Framework skill, make a cheeky REST call or dive into MCP (currently public preview), pick the right tool for the job and your copilot will sing. Ready to give it a whirl? Onwards!

Sign up for more like this.

Enter your email
Subscribe

Disclaimer: The software, source code and guidance on this website is provided "AS IS"
with no warranties of any kind. The entire risk arising out of the use or
performance of the software and source code is with you.

Any views expressed in this blog are those of the individual and may not necessarily reflect the views of any organization the individual may be affiliated with.