D365 F&O MCP Server

GitHub Copilot excels at C#, Python, and JavaScript — languages with rich public training data. X++ is different. Your D365FO codebase is private, highly customized, and deeply interconnected: thousands of CoC extensions layered over standard Microsoft code, ISV packages adding their own tables and classes, custom EDTs that drive field validation across dozens of forms. Copilot has never seen any of it.

The result is AI that confidently generates code that doesn't compile: wrong method signatures, missing parameters, fields that don't exist on the table, CoC chains broken because Copilot didn't know an extension already wrapped the method.

This MCP server solves that by pre-indexing your entire D365FO installation — 584,799+ symbols across standard and custom models — and exposing it to Copilot as 54 specialized tools. Before generating any X++ code, Copilot can look up exact method signatures, check what CoC extensions already exist, trace security hierarchies, find label translations, and understand the full shape of your data model. The result is code that compiles on the first try and integrates correctly with your existing customizations.

• Massive Metadata Index: Instantly looks up signatures, tables, enums, EDTs across standard and ISV code out of 580,000+ objects.
• Smart Object Generation: AI-driven tools build XML structures exactly matching D365 standard patterns (SimpleList, Forms, Tables).
• X++ Knowledge Base: Queryable knowledge base of D365FO patterns, best practices, and AX2012→D365FO migration guidance — prevents deprecated API usage.
• Code Review & Git Diff: Reviews uncommitted workspace changes locally matching D365 Best Practice metrics directly against Copilot chat.
• D365 SDLC Native Integration: Triggers local MSBuild, sync.exe database updates, SysTestRunner tests, and xppbp.exe best practice validations smoothly behind the scenes.
• XML Parsing without Corruption: Intelligently mutates AxTable and AxForm files locally avoiding the common string replacement corruption associated with VS agents.

Full step-by-step guide with all scenarios: docs/QUICK_START.md

Prerequisites — install required software via d365fo.tools:

Install-D365SupportingSoftware -Name vscode,python,node.js

git clone https://github.com/dynamics365ninja/d365fo-mcp-server.git
cd d365fo-mcp-server
npm install

# Build the C# bridge (required on Windows D365FO VMs for file create/modify)
cd bridge\D365MetadataBridge
dotnet build -c Release
cd ..\..                         # Back to repo root

copy .env.example .env           # Set PACKAGES_PATH, CUSTOM_MODELS, LABEL_LANGUAGES, ...
npm run extract-metadata         # Extract XML from D365FO packages
npm run build-database           # Build SQLite index
npm run dev

UDE / Power Platform Tools? Run npm run select-config instead of setting PACKAGES_PATH manually.

1. Enable MCP servers in Copilot at github.com/settings/copilot/features

2. In Visual Studio: Tools → Options → GitHub → Copilot → check Enable MCP server integration in agent mode

3. Create %USERPROFILE%\.mcp.json (covers all solutions on the machine, recommended) or place .mcp.json next to a specific .sln file:

{
  "servers": {
    "d365fo-azure": {
      "url": "https://your-server.azurewebsites.net/mcp/"
    },
    "d365fo-local": {
      "command": "node",
      "args": ["K:\\d365fo-mcp-server\\dist\\index.js"],
      "env": {
        "MCP_SERVER_MODE": "write-only",
        "D365FO_SOLUTIONS_PATH": "K:\\VSProjects\\MySolution",
        "D365FO_WORKSPACE_PATH": "K:\\AosService\\PackagesLocalDirectory\\YourPackageName\\YourModelName"
      }
    }
  }
}

4. Copy the Copilot instruction files so Copilot knows the D365FO workflow rules:

# Place .github in a parent folder shared by all your D365FO solutions, e.g.:
Copy-Item -Path ".github" -Destination "C:\source\repos\" -Recurse

Tip: Visual Studio 2022 searches for .github\copilot-instructions.md upward from the solution folder, so one copy in a common parent directory covers all solutions underneath — no need to copy it into every solution separately.

Full config options (UDE paths, explicit projectPath, solutionPath): docs/MCP_CONFIG.md

Host on Azure App Service so the whole team shares one instance — nobody needs the server running locally.

The database downloads from Azure Blob Storage automatically on startup.

Setup guide: docs/SETUP.md · CI/CD pipeline: docs/PIPELINES.md