Power BI Analyst MCP

Ask Claude to analyse your Power BI data. Get answers — not context-window crashes.

Connect Claude (or any MCP client) to your Power BI semantic models. Explore tables and measures, run DAX queries, and work with real results — even across datasets with tens of thousands of rows. Large query results are automatically saved to a local file and paged to the agent on demand, so your AI session stays fast and focused no matter how much data you pull.

Everything runs on your machine. Your data never passes through a third-party relay.

What becomes possible

The analytics bottleneck has never really been data access — Power BI already gives people access. The bottleneck is translation: the skilled, time-consuming work of turning data into a decision. This server moves that translation to an LLM.

Compound reasoning across your entire model A human analyst runs one query, reads the result, forms a hypothesis, runs another query. This serialises over hours. An agent can run twenty queries in sequence — each informed by the last — synthesise across all of them, and deliver a reasoned conclusion in minutes. Ask "what's driving the margin decline in EMEA?" and Claude will explore measures, drill into markets, check time trends, isolate the outlier, and explain it — without you directing each step.

Natural language analytics for everyone Any stakeholder can ask a data question and get a real answer backed by live DAX — without knowing what DAX is, without filing a ticket, without waiting. The translation layer that used to require a trained analyst runs on demand.

Proactive anomaly detection Run Claude on a schedule against your key measures. It queries the data, compares to prior periods, and flags anything outside expected ranges in plain English — before anyone has to open a dashboard to find out something went wrong.

Self-documenting semantic models "List every measure in this dataset and explain what it calculates." Claude explores the schema and produces a data dictionary — useful for onboarding, governance, and anyone trying to understand what a model actually contains.

Large datasets, handled automatically Power BI queries can return tens of thousands of rows. Returning all of that inline would consume most of an LLM's context window and crash the session. This server saves large results to a local CSV and gives the agent a compact summary — row count, column names, 5-row preview — then lets it page through the file on demand. You get the full dataset. The AI session stays lean.

Query history that compounds over time Every successful DAX query is logged locally in a JSONL audit trail — what the user asked, the DAX that was generated, the columns returned, and where the CSV was saved. In the next session, the agent searches this history to find relevant prior work: reusable DAX patterns, previously computed result files, and context from earlier analyses. The more you use it, the faster and smarter it gets — and you always have an audit trail of where every number came from.

Your data stays on your machine Queries, results, and tokens never pass through a cloud relay. The server runs locally, authenticates via the same OAuth device code flow as the Power BI web app, and stores tokens in your OS's native secure store (Keychain / DPAPI / LibSecret).

How it compares to Microsoft's official MCP servers

Microsoft publishes two MCP servers for Power BI. This one is different in purpose and architecture from both.

What it looks like

You: Analyse revenue by market and product category for Q1 2025

Claude: Let me explore the dataset first.
        [calls list_tables → list_measures → execute_dax]

        The query returned 73,840 rows — saved to:
        ~/powerbi_output/dax_result_revenue_q1_2025_20260313_091204.csv

        [pages through results with read_query_result]

        Summary: EMEA leads at 44% of total revenue. The top category
        is Premium Hardware in both EMEA and AMER. APAC shows the
        strongest quarter-over-quarter growth at +18%...

The agent explores the schema, writes the DAX, handles the file, and delivers the analysis — without you touching the Power BI UI.

Tools

User guide

Prerequisites

• Python 3.10+
• A Power BI Pro, Premium Per User (PPU), or Premium capacity licence
• An Azure AD app registration (free, ~5 minutes — see below)

Step 1 — Create an Azure AD app registration

OAuth 2.0 requires a client ID to identify which application is acting on your behalf. The registration is free, requires no client secret, and does not need Power BI admin consent for the two read-only scopes used here.

1. Go to portal.azure.com → Azure Active Directory → App registrations → New registration.

2. Name it (e.g. PowerBI MCP). For account types choose Accounts in this organizational directory only (single-tenant). Click Register.

3. Under Authentication → Platform configurations, add Mobile and desktop applications and tick this redirect URI:

   Code

   Copy

   https://login.microsoftonline.com/common/oauth2/nativeclient
   

4. Still under Authentication → Advanced settings, set Allow public client flows to Yes. Save.

5. Under API permissions → Add a permission → Power BI Service, add:

   • Dataset.Read.All
   • Workspace.Read.All

   If your tenant requires admin consent, ask an admin to grant it.

6. From the Overview page, copy both of these — you will need them in Step 2:

   • Application (client) ID
   • Directory (tenant) ID

Note: The Power BI tenant setting "Dataset Execute Queries REST API" must be enabled in the Power BI Admin Portal (Integration settings) for execute_dax to work.

Step 2 — Install and connect

The server is published on PyPI. The fastest way to run it is with uvx, which requires no manual install or virtual environment.

Claude Desktop

Add the following to your claude_desktop_config.json:

{
  "mcpServers": {
    "powerbi": {
      "command": "uvx",
      "args": ["powerbi-analyst-mcp"],
      "env": {
        "POWERBI_CLIENT_ID": "your-application-client-id",
        "POWERBI_TENANT_ID": "your-directory-tenant-id"
      }
    }
  }
}

The config file is at:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

POWERBI_TENANT_ID is required for almost all users. Most organisations have a single Azure AD tenant. Set this to your Directory (tenant) ID from Step 1. Leaving it as organizations (the default) will cause authentication to fail or target the wrong tenant.

Cursor

Add a .cursor/mcp.json file in your project (or use the global config):

{
  "mcpServers": {
    "powerbi": {
      "command": "uvx",
      "args": ["powerbi-analyst-mcp"],
      "env": {
        "POWERBI_CLIENT_ID": "your-application-client-id",
        "POWERBI_TENANT_ID": "your-directory-tenant-id"
      }
    }
  }
}