Let Any Agent Explore Power Platform Solutions via Chat

Watch the session

About this session

Power Platform solutions can contain hundreds of components - Cloud Flows, Copilot Studio agents, Dataverse tables, custom connectors, environment variables, canvas apps, and more. Understanding what is inside one typically means navigating the maker portal or reading XML files by hand. This session introduces a different model: a Copilot Studio agent wired to the open-source Power Platform Solution Explorer MCP server can inspect, list, and search any solution's contents in plain language. Export a solution ZIP, start the server, connect the agent - and your AI can answer questions about the solution without you ever opening a file.

Inside the demo

• Export a Power Platform solution ZIP from the maker portal and drop it into the server's watch folder - no other configuration required.
• Connect the MCP server to a Copilot Studio agent via Tools > Add tool > MCP, using the /mcp endpoint and an API key header.
• Ask: 'What components are in this solution?' - the agent calls list_components and returns a structured breakdown across 15 component types.
• Ask: 'Show me the system prompt for the Copilot Studio agent in this solution.' - get_component_detail drills into the bot component and surfaces the full instruction set.
• Ask: 'Which flows are in this solution and what connectors do they use?' - the agent surfaces flow definitions with connection references.
• Ask: 'Search the entire solution for any reference to the word invoice.' - search_solution scans every file in the solution.
• load_solution_context returns the complete solution context in one call - enough for the agent to answer most questions without follow-up tool calls.

Prompts to try

Point an agent at your own Process Mining data and adapt these:

• ›What components are in this solution?
• ›Show me the system prompt for the Copilot Studio agent.
• ›Which flows are in this solution and what connectors do they use?
• ›List all Dataverse tables with their columns and relationships.
• ›Search the entire solution for any reference to 'invoice'.
• ›What environment variables does this solution define?

Build it yourself

1. 1

   Export your solution

   From the Power Platform maker portal, export any solution as an unmanaged ZIP. Drop it into the server's configured watch folder - the server auto-extracts it on start.

   Export solutions (Microsoft Learn)→
2. 2

   Clone and start the MCP server

   Clone the GitHub repo, install dependencies with npm install, add your API key to .env, and run npm start. The server listens on :3001/mcp by default.

   powerplatform-solution-mcp on GitHub→
3. 3

   Connect to Copilot Studio

   In Copilot Studio, go to Tools > Add tool > MCP. Enter your server's public /mcp endpoint URL. Set authentication to Header type with your API key header name and value.

   Connect an MCP server to Copilot Studio→
4. 4

   Enable generative orchestration

   Generative orchestration is required for the agent to choose which of the 9 tools to call at runtime based on the user's question. Enable it in your agent's settings before testing.

   Generative orchestration→
5. 5

   Test and explore

   Ask the agent about your solution's components, flows, and Dataverse schema. Use load_solution_context when you want the agent to have full context before answering a complex question.

   All 9 tools documented in the README→

Resources