Power Platform Solution Explorer MCP
An open-source MCP server with 9 tools that lets AI agents like Copilot Studio deeply inspect Power Platform solution files.
Project details
What It Does
Solution Explorer MCP is a lightweight server that reads exported Power Platform solution ZIP files and exposes their contents through the Model Context Protocol (MCP). This lets any MCP-compatible AI agent, including Microsoft Copilot Studio, inspect and reason about the full structure of a solution without manual exploration.
Drop a .zip solution export into a folder, start the server, and your AI agent instantly has access to 9 consolidated tools. Use list_components with a component type enum to browse any of 15 component types, and get_component_detail to drill into flows, entities, or bot components.
Why It Matters
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 a solution typically requires navigating the maker portal or manually reading XML and JSON files.
With this MCP server, an AI agent can answer questions like:
- "What flows are in this solution and what do they connect to?"
- "Show me the system prompt for the Copilot Studio agent."
- "List all Dataverse tables with their columns and relationships."
- "Search every file in the solution for a specific keyword."
This turns solution analysis from a manual task into a conversational experience.
Architecture
The server follows a stateless factory pattern using Streamable HTTP transport, making it compatible with Copilot Studio's MCP connector (currently in preview). Each request creates a fresh MCP server instance. No database, no sessions, no cleanup required.
ZIP drop folder
|
v
[Auto-extract] --> [Extracted solutions]
|
v
[MCP Server :3001/mcp]
|
|-- 9 Tools
| (list_components, get_component_detail, search_solution, ...)
|
|-- Dynamic Resources
| (solution overviews, flow definitions, agent prompts)
|
|-- API Key Auth
| (configurable header-based authentication)
|
v
[Copilot Studio / Any MCP Client]The 9 Tools
Each tool takes a solution folder name and returns structured JSON. The consolidated set covers:
| Category | Tools |
|---|---|
| Solution | list_solutions, get_solution_info, get_solution_structure, load_solution_context |
| Components | list_components (15 component types via enum), get_component_detail (flow, entity, bot_component) |
| Utility | search_solution, get_raw_file, get_sitemap |
Copilot Studio Integration
Connecting to Copilot Studio takes 3 steps:
- Start the server and expose it via a public URL (e.g., using a tunnel or cloud deployment).
- In Copilot Studio, go to Tools > Add tool > MCP and enter the
/mcpendpoint URL. - Configure authentication with Header type, entering the API key header name and value.
Once connected, the agent can use any of the 9 tools directly in conversation. The load_solution_context tool is especially powerful as it returns the complete solution context in a single call, giving the agent enough information to answer most questions without follow-up tool calls.
Key Features
- Zero config start: Drop a ZIP, run the server, connect your agent.
- API key authentication: Configurable header name and key via environment variables.
- Dataverse schema parsing: Columns with types, 1:N and N:N relationships, entity keys, forms, views, and charts.
- Knowledge source detection: Parses Copilot Studio knowledge configurations (SharePoint sites, Dataverse search, uploaded files).
- Full-text search: Search across every file in a solution for any keyword.
- Dynamic MCP resources: Solution overviews, flow definitions, and agent prompts available as context resources.
- Stateless design: Each request creates a fresh MCP server instance.
Project Repository
Solution Explorer MCP is open source. The complete TypeScript implementation, including all 9 tools, XML/JSON parsers, and the HTTP server, is available on GitHub.
