Understanding SharePoint Copilot Apps

[siddharth-vaghasia]

What type of issue is this?

Question

What SharePoint development model, framework, SDK or API is this about?

💥 SharePoint Framework

Target SharePoint environment

SharePoint Online

What browser(s) / client(s) have you tested

• 💥 Internet Explorer
• 💥 Microsoft Edge
• 💥 Google Chrome
• 💥 FireFox
• 💥 Safari
• mobile (iOS/iPadOS)
• mobile (Android)
• not applicable
• other (enter in the "Additional environment details" area below)

Additional environment details

NA

Issue description

So trying to understand Copilot Components, below are some questions

1. How is Copilot Component related to Copilot Agent(Declartive) ? I understand one Agent can have multiple components which can be called based on user intent?
2. Each Copilot component can have multiple tools in its manifest, trying to understand, is this to support Single React component can render diffrent UI based on the tool being called ? but which parameter we can use to control this in properties ? Will this always be based on the message parameter that our code has to understand and then redner conditional User interface?
3. Is it necessary that when we are buidling Copilot Apps or say Component, it needs to have its own Declartive Agent ? Can Microsoft 365 not call our Copilot component in a tenant if it is deployed globally ?

[jansenbe]

@siddharth-vaghasia Here's some answers to your questions:

1. SharePoint Copilot Apps are deployed as declarative agent. You can fully extend the declarative agent that's included in the SPFx solution. Every SharePoint Copilot App has one or more Copilot Components which are mapped to one or more tool calls. These tool calls and up as MCP tools returning the respective Copilot Component after deployment. This is based on the MCP App support in Copilot. When creating the components with their tools you can set the needed parameters and descriptions for the AI backend to understand which tool call to make and what input to feed in

2. Correct, every component can have multiple tool calls and you can have multiple components in a single solution, so you can end up with a DA that returns different UX components depending on the use ask. At the moment BaseCopilotComponent does not surface toolName on this.properties, but that's something we'll fix

3. At the moment yes, Declarative Agents is the mechanism to use SharePoint Copilot Apps in Copilot. Both we and Copilot are working on making Copilot components easier to consume.

[siddharth-vaghasia]

thanks Bert, appreicate your prompt response... couple of follow up quesitons

[Bert] - When creating the components with their tools you can set the needed parameters and descriptions for the AI backend to understand which tool call to make and what input to feed in
[Sid] - So for every component we hvae a manifest.json file which has tools as array, so here we are saying we can add multiple tools and use description property to let AI know which tool to be called. Is this the only place to let AI know which tools to be called by giving description here?

The documentation says Multiple tools in one package, in that context, should we also reflect a text here saying a Copilot component can also have multiple tools... ?

I understand you said the toolName we will get in this.properties will be added later which will help us do conditional UX rendering....
but is there any way to get use the actual user message text which triggered this tool ?

[jansenbe]

@siddharth-vaghasia : yes, a copilot component can have multiple tools. We don't have a sample for that yet but will paste some code below. Also, check out https://github.com/pnp/spfx-copilot-apps/tree/main/samples/time-off-absence/src/copilotComponents to learn more about how to set tool descriptions: A) you've the tool description in the manifest:

B) you've the parameter descriptions in the *properties.ts file:

Here's a sample of multiple tools with properties for the same component. Manifest:

and the properties.ts file

[jansenbe]

And to answer your last question @siddharth-vaghasia , you can infer the tool called from it's parameters you receive as properties. Below code shows this. This is something we'll improve before GA

[siddharth-vaghasia]

thanks @jansenbe for all the responses, this helped me a lot to understand it better, have build a small sample, will add it to our sample gallery soon.!!!