How Exagomatica connects Claude to your data stack.
Exagomatica is a local MCP server that connects Claude Desktop to your data over stdio. It gives Claude strict read-only access to your Power BI semantic model and the pipelines that feed it, adds optional change tracking from Azure DevOps, and respects your existing authentication and authorisation.
- Self-contained Windows app
- MCP over stdio
- One Microsoft sign-in
- Read-only
One local process between Claude and your data.
Claude Desktop launches Exagomatica as a subprocess on first connection. All communication happens over stdio: no network ports opened, no inbound listener. The server then makes outbound REST calls to enterprise APIs using the user’s own Microsoft (Entra ID) sign-in.
AI client
Claude Desktop
Natural language
MCP / stdio
MCP server
Exagomatica
Read-only tools · auth · DAX validator · ETL KB
REST · Entra ID
Power BI
Semantic model + dataflows
Azure DevOps
Change tracking · optional
SharePoint
via M365 connector
SharePoint document search is provided by Anthropic’s official Microsoft 365 connector in Claude Desktop (Settings › Connectors). Exagomatica’s document-search skill orchestrates whichever SharePoint search and read capabilities the connector exposes at runtime.
Read-only capabilities across four domains.
These are the capabilities Claude draws on to answer you; you never invoke them yourself. Every data capability is read-only and respects existing access controls, and a licensing check is always available so Claude can report your license state.
Power BI Intelligence
Get Semantic Model Schema
Returns tables, columns, measures, and relationships. Claude uses it to understand what data is available before writing a query.
Execute DAX Query
Runs DAX EVALUATE statements against live data. Claude pre-validates names against the schema, checks syntax before execution, and corrects itself. Results include an interactive chart view.
Get Measure Definition
Returns the DAX formula and metadata for a specific measure. Claude uses it to explain how a KPI is computed.
Architecture & Knowledge Base
Architecture Overview
Renders the unified data architecture: semantic-model tables/columns/measures on top, ETL entities in the middle, external sources at the bottom, with lineage edges connecting layers.
Search Architectural Entities
Finds where transformations happen by entity name. Searches the model’s own Power Query, dataflow entities read live from the Power BI service, and, when change tracking is connected, git-tracked queries and partitions.
Get Architectural Entity Details
Returns the full transformation pipeline with business logic annotations, partition breakdown, and optionally the complete M code.
List All Sources
Lists every source backing the architectural index: the Power BI model itself, each Service dataflow it links to, and, when change tracking is connected, each Azure DevOps repository.
Refresh Knowledge Base
Builds the knowledge base from every configured source: the model’s own Power Query, the Service dataflows it links to, and, when connected, Azure DevOps repos. Supports full, incremental, and entity-specific refreshes. Auto-triggers on first use each session.
Get Knowledge Base Statistics
Returns entity counts, source coverage, semantic-model snapshot freshness, and last scan time.
DAX Reference
Look Up DAX Function
Retrieves syntax, parameters, and examples for any DAX function from dax.guide. Helps Claude write idiomatic DAX.
Licensing
Get License Status
Reports your active tier, expiry, and feature flags. Claude can always check it, even when licensing blocks the data capabilities.
Pre-built workflows.
Skills guide Claude through multi-step tasks, ensuring the right capabilities are used in the right order with appropriate context. You ask in plain English; Claude picks the workflow.
Analyze data from Power BI
Guided ad-hoc analysis. Discovers schema, writes optimised DAX, returns charts, and enriches answers with ETL pipeline context.
Trace data lineage
Traces data from report columns back to source through every transformation step. Shows how data is filtered, renamed, and enriched.
Investigate unexpected data
Answers “why is this number wrong, zero, or missing?” Picks the likely cause across measure, lineage, and calculation and returns a grounded hypothesis you can act on.
Explain a Power BI measure
Retrieves a measure’s DAX formula and translates it into plain business language, grounded in the upstream data that feeds it, so non-technical users understand exactly what the number means.
Top / Bottom N analysis
Quick ranked analysis. Returns the top or bottom N records by any metric. Great for leaderboards and outlier detection.
Find business documentation
Searches SharePoint for business docs and glossaries via the Microsoft 365 connector. Summarises findings so users get answers without reading whole documents.
Refresh the ETL knowledge base
Rebuilds the local knowledge base from every configured source: the model’s own Power Query, its linked dataflows, and, when connected, Azure DevOps repos. Useful after pipeline or model changes.
Designed for enterprise environments.
Exagomatica is designed for environments where data security is non-negotiable.
One Microsoft sign-in
Sign in once during setup with your Microsoft (Entra ID) account, MFA included. The server refreshes tokens silently; no separate credentials to manage.
Local execution
The server runs on the user’s machine, read-only. It adds no cloud of its own to your data path: data flows between your enterprise APIs and Claude Desktop, with no telemetry.
Permission-aware
On a standard install, every query runs as the signed-in user, so Power BI applies its own rules: row-level security for Viewer-role users, and the wider view that workspace Contributors already have. Users see exactly what Power BI lets them see.
- Read-only access, no DDL operations
- DDL blocklist enforced at the server level
- Sign-in tokens encrypted at rest (Windows DPAPI)
- No telemetry or data collection
- License key via environment variable
- Per-query result caps (default 1,000 rows, max 10,000)
- Full audit trail in local logs
- MCP stdio transport, no open network ports
A note on what “local” means: Claude Desktop itself sends your questions (and the results Exagomatica returns) to Anthropic’s API to generate answers, the same as any AI assistant. “Local” means Exagomatica adds no cloud of its own to your data path, not that the assistant runs offline.
Technical specifications.
- Runtime
- Self-contained, no Python install required
- MCP SDK
- mcp (FastMCP)
- Protocol
- Model Context Protocol over stdio
- Authentication
- Microsoft Entra ID (MSAL), one sign-in
- AI client
- Claude Desktop
- Platforms
- Windows 10/11
- Installer
- Windows installer (Inno Setup)
- Data access
- Read-only, DDL blocked at server level