🔌 Dataverse MCP Server

A Model Context Protocol (MCP) server for Microsoft Dataverse, enabling AI assistants like Claude to interact with Dataverse data through natural language.

Built on: Model Context Protocol | Microsoft Dataverse MCP Spec

What is This?

This MCP server exposes Microsoft Dataverse tables and records through the Model Context Protocol, allowing AI assistants to:

📊 List and explore Dataverse tables (entities)
🔍 Query records with OData filters and sorting
✏️ Create new records in any table
🔄 Update existing records by GUID
🤖 Natural language interface via Claude Desktop, VS Code, or any MCP client

Quick Start

1. Prerequisites

Python 3.11+
Dataverse environment with API access
Azure AD app registration with Dataverse permissions

2. Install

# Install dependencies pip install fastapi uvicorn requests python-dotenv fastmcp # Or with uv (recommended) uv pip install fastapi uvicorn requests python-dotenv fastmcp

3. Configure

Run the setup script:

chmod +x setup_dataverse.sh ./setup_dataverse.sh

Or manually create .env.local:

DATAVERSE_HOST=https://org1bfe9c69.api.crm.dynamics.com DATAVERSE_TENANT_ID=your-tenant-id DATAVERSE_CLIENT_ID=your-client-id DATAVERSE_CLIENT_SECRET=your-client-secret

4. Test Connection

python test_dataverse.py

Expected output: ✅ All tests passed! Dataverse MCP server is ready.

5. Start Server

Option A: Run Locally

./watch.sh # Or: uvicorn server.app:combined_app --reload --port 8000

Option B: Deploy to Databricks Apps

./deploy.sh --create

📖 See DATABRICKS_DEPLOYMENT.md for deployment guide.

Server endpoints:

🔧 MCP: http://localhost:8000/mcp (local) or https://your-app.databricksapps.com/apps/your-app/mcp (cloud)
📖 Docs: http://localhost:8000/docs
❤️ Health: http://localhost:8000/api/health

Documentation

Document	Description
DATAVERSE_QUICKSTART.md	5-minute quick start guide
DATAVERSE_SETUP.md	Detailed Azure AD & Dataverse setup
DATABRICKS_DEPLOYMENT.md	Deploy to Databricks Apps
README_DATAVERSE.md	Complete reference documentation

Available MCP Tools

Phase 1 (Implemented)

Tool	Description
`health`	Server health check
`list_tables`	List all Dataverse tables
`describe_table`	Get table schema and columns
`read_query`	Query records with OData filters
`create_record`	Create new records
`update_record`	Update existing records

Phase 2 (Planned)

delete_record - Delete records
create_table / update_table / delete_table - Table management
list_knowledge_sources / retrieve_knowledge - Copilot Studio integration
list_prompts / execute_prompt - Custom prompts

Usage Examples

Claude Desktop

Configure ~/Library/Application Support/Claude/claude_desktop_config.json:

{ "mcpServers": { "dataverse": { "url": "http://localhost:8000/mcp" } } }

Then in Claude:

You: "List all custom tables in my Dataverse environment" Claude: [Calls list_tables with custom_only=True] You: "Show me accounts with revenue over $1M" Claude: [Calls read_query with filter] You: "Create a new account called Fabrikam with $2M revenue" Claude: [Calls create_record]

cURL

# List tables curl -X POST http://localhost:8000/mcp/call-tool \ -H "Content-Type: application/json" \ -d '{"name": "list_tables", "arguments": {"top": 10}}' # Query accounts curl -X POST http://localhost:8000/mcp/call-tool \ -H "Content-Type: application/json" \ -d '{ "name": "read_query", "arguments": { "table_name": "account", "select": ["name", "revenue"], "filter_query": "revenue gt 1000000", "top": 10 } }'

Python

import requests response = requests.post('http://localhost:8000/mcp/call-tool', json={ "name": "describe_table", "arguments": {"table_name": "account"} }) print(response.json())

Project Structure

dataverse_mcp_server/ ├── server/ │ ├── app.py # FastAPI + MCP server │ ├── dataverse_tools.py # MCP tool implementations │ ├── dataverse/ │ │ ├── auth.py # OAuth authentication │ │ ├── client.py # Dataverse Web API client │ │ └── __init__.py │ └── routers/ # FastAPI REST endpoints │ ├── health.py # Health check │ ├── mcp_info.py # MCP metadata │ └── user.py # User info ├── client/ # React frontend (future) ├── test_dataverse.py # Connection tests ├── setup_dataverse.sh # Interactive setup ├── watch.sh # Dev server with hot reload ├── config.yaml # MCP server config ├── env.example # Environment template ├── DATAVERSE_QUICKSTART.md # Quick start guide ├── DATAVERSE_SETUP.md # Detailed setup └── README_DATAVERSE.md # Full reference

Architecture

Authentication: Service Principal (OAuth 2.0 client credentials flow)
API: Dataverse Web API v9.2
Protocol: Model Context Protocol (MCP)
Framework: FastAPI + FastMCP

MCP Clients (Claude, VS Code) ↓ HTTP/MCP Dataverse MCP Server (FastAPI) ↓ HTTPS/OAuth Microsoft Dataverse

Troubleshooting

Problem	Solution
401 Authentication Error	Check client secret validity and API permissions
403 Permission Denied	Assign security role to app user in Power Platform Admin Center
Table Not Found	Use `list_tables()` to find correct logical name
Connection Timeout	Verify `DATAVERSE_HOST` and network connectivity

See DATAVERSE_SETUP.md for detailed troubleshooting.

Development

# Run tests python test_dataverse.py # Start dev server ./watch.sh # Check lints ruff check server/ # Format code ruff format server/

Reference Links

License

See LICENSE.md

Security

Report vulnerabilities via SECURITY.md

Questions? See DATAVERSE_QUICKSTART.md or DATAVERSE_SETUP.md

Built with ❤️ for the Dataverse and MCP communities

API Registry MCP Server