# Pbix Mcp

> MCP server for full read/write access to every layer of Power BI .pbix files — layout, DAX measures, M code, metadata SQL, VertiPaq table data, themes, and more.

- **Type:** MCP server
- **Install:** `agentstack add mcp-d0nk3yhm-pbix-mcp`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [d0nk3yhm](https://agentstack.voostack.com/s/d0nk3yhm)
- **Installs:** 0
- **Category:** [Databases](https://agentstack.voostack.com/c/databases)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [d0nk3yhm](https://github.com/d0nk3yhm)
- **Source:** https://github.com/d0nk3yhm/pbix-mcp

## Install

```sh
agentstack add mcp-d0nk3yhm-pbix-mcp
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# pbix-mcp

[](https://github.com/d0nk3yhm/pbix-mcp/actions/workflows/ci.yml)
[](https://pypi.org/project/pbix-mcp/)
[](https://pypi.org/project/pbix-mcp/)
[](https://www.python.org/downloads/)
[](https://opensource.org/licenses/MIT)

An MCP server for **creating**, reading, writing, and evaluating Power BI `.pbix` and `.pbit` files — **no Power BI Desktop required**. The entire PBIX binary format has been independently reversed and reimplemented in pure Python — no templates, no skeletons, no Microsoft binaries. Generated files open in PBI Desktop with full interactivity: view data, add measures, create visuals, and refresh — verified with PBI Desktop March 2026.

Exposes 101 tools covering report creation (all 6 data types, cross-table relationships, CSV/SQLite/SQL Server/MySQL/PostgreSQL/Excel/JSON/Azure SQL data sources, DirectQuery, and DAX measures), layout editing, visual management, bookmarks, custom visuals, field parameters, calculation groups, TMDL export, incremental refresh, DAX evaluation (156 functions), RLS security, and binary format internals.

See [CHANGELOG.md](CHANGELOG.md) for version history.

## Try It

Generate a complete 3-page Northwind Analytics Dashboard in under a second:

```bash
pip install pbix-mcp
python examples/create_showcase.py
```

Creates a PBIX with 6 tables, 5 relationships (including chained cross-table lookups), 4 DAX measures, 3 pages, and 14 visuals. Open `showcase_northwind.pbix` in Power BI Desktop — everything works: slicers, cross-filtering, drill-through.

## Quick Start

```bash
git clone https://github.com/d0nk3yhm/pbix-mcp.git
cd pbix-mcp
pip install -e .
```

### Claude Desktop / Claude Code

Add to your MCP config file:

| Platform | Config file |
|----------|------------|
| Claude Desktop (macOS) | `~/Library/Application Support/Claude/claude_desktop_config.json` |
| Claude Desktop (Windows) | `%APPDATA%\Claude\claude_desktop_config.json` |
| Claude Code | `~/.claude/settings.json` (Linux/macOS) or `%USERPROFILE%\.claude\settings.json` (Windows) |

```json
{
  "mcpServers": {
    "powerbi-editor": {
      "command": "pbix-mcp-server"
    }
  }
}
```

> **Windows note:** If `pbix-mcp-server` is not on PATH, use the full Python path:
> ```json
> {
>   "mcpServers": {
>     "powerbi-editor": {
>       "command": "python",
>       "args": ["-m", "pbix_mcp.cli"]
>     }
>   }
> }
> ```

### Codex Desktop (OpenAI)

1. `pip install git+https://github.com/d0nk3yhm/pbix-mcp.git`
2. Open Codex Desktop → Settings → MCP → Add Server
3. Configure:
   - **Name**: `powerbi-editor`
   - **Command**: `pbix-mcp-server`
   - **Arguments**: *(leave empty)*

### Generic MCP (stdio)

```bash
pbix-mcp-server
# With debug logging:
pbix-mcp-server --log-level debug
```

## Format Reversal Status

Every layer of the PBIX binary format has been independently reversed and reimplemented. No templates, skeletons, or Microsoft binaries are used.

| Layer | Status | Implementation |
|-------|--------|----------------|
| PBIX ZIP shell | **Reversed** | Version, Content_Types, DiagramLayout, Settings, Metadata — generated constants |
| Report/Layout JSON | **Reversed** | Pages, visuals, data bindings, filters — `_build_layout()` |
| ABF binary container | **Reversed** | 72-byte signature, BackupLogHeader, VirtualDirectory, BackupLog — `build_abf_clean()` |
| XMLA Load document (db.xml) | **Reversed** | 28 xmlns namespaces, CompatibilityLevel=1550, TabularMetadata — `generate_db_xml()` |
| CryptKey.bin | **Constant** | 144-byte RSA key BLOB (Microsoft crypto format; GUID-independent constant) |
| Metadata SQLite | **Reversed** | 68 system tables matching PBI March 2026 schema — `create_empty_metadata_db()` |
| VertiPaq column storage | **Reversed** | IDF (bit-packed), IDFMETA (segment stats), dictionary (Long/Real/String), HIDX (hash index) |
| H$ attribute hierarchies | **Reversed** | NoSplit POS_TO_ID + ID_TO_POS for all cardinalities |
| R$ relationship indexes | **Reversed** | NoSplit INDEX encoding with +3 DATA_ID_OFFSET padding and 1-based row indices (verified byte-exact against PBI Desktop ground truth) |
| XPress9 compression | **Reversed** | Custom compress/decompress with reversed chunk framing, headers, and multi-thread format; core algorithm via [xpress9-python](https://github.com/Hugoberry/xpress9-python) (MIT) |

The only non-generated artifact is the 144-byte CryptKey constant. This is a Microsoft RSA key BLOB that requires `rskeymgmt` infrastructure to generate. The key is GUID-independent — any valid key works with any database ID. Random bytes produce `PFE_INVALID_CRYPT_KEY`.

## Stability

| Feature | Status | Notes |
|---------|--------|-------|
| PBIX creation | **Stable** | Multi-table with all 6 data types, relationships, H$ hierarchies, and measures. Generated files support full PBI Desktop editing (add measures, columns, visuals) |
| Cross-table relationships | **Stable** | R$ system tables with NoSplit INDEX encoding (+3 padding, 1-based row indices); cross-table visuals, RELATED(), and cross-table filtering verified byte-exact against PBI Desktop ground truth |
| Refreshable CSV sources | **Stable** | `source_csv` parameter creates M expressions referencing external CSV files; click Refresh in PBI Desktop to re-import |
| SQLite database sources | **Stable** | `source_db` with ODBC driver; data imported at build, Refresh re-reads from DB |
| SQL Server / MySQL / PostgreSQL database sources | **Stable** | `source_db` Import and DirectQuery for all. MySQL DQ requires MariaDB ODBC 3.1 (`type: 'mariadb'`) |
| Excel data sources | **Stable** | `source_db` with `type: 'excel'` — Import mode |
| JSON/API data sources | **Stable** | `source_db` with `type: 'json'` — Import mode from REST APIs and JSON files |
| Azure SQL data sources | **Stable** | `source_db` with `type: 'azuresql'` — Import and DirectQuery |
| Data source switching | **Stable** | `pbix_update_data_source` — lightweight connection string change without full DataModel rebuild. Switch between SQL Server, PostgreSQL, MySQL, CSV, Excel, JSON, SQLite, Azure SQL. Switch Import/DirectQuery mode. Verified with live MSSQL→PostgreSQL→CSV roundtrip |
| DirectQuery mode | **Stable** | `mode='directquery'` with SQL Server, PostgreSQL, and MySQL (via MariaDB ODBC 3.1) — live database queries, no refresh needed |
| VertiPaq table data write | **Stable** | Create and roundtrip (set_table_data, update_table_rows, replace_value) via full builder rebuild |
| Data export | **Stable** | `pbix_export_table_csv`, `pbix_export_all_tables_csv` — export any table(s) to CSV, all rows, proper quoting, ISO dates |
| Data search & query | **Stable** | `pbix_find_value` across tables, `pbix_query_table` SQL-like WHERE/AND/OR/LIKE/IN/ORDER BY, `pbix_table_stats` profiling, `pbix_data_diff` row-level file comparison |
| Roundtrip DataModel modify | **Stable** | Add/remove tables, relationships, measures on existing files. Metadata-only changes (measures, RLS, column properties) use binary splice for PBI Desktop files; structural changes use full builder rebuild |
| H$ attribute hierarchies | **Stable** | NoSplit POS_TO_ID + ID_TO_POS for all cardinalities; MaterializationType=0 |
| Report layout read/write | **Stable** | Pages, visuals, filters, positions, bookmarks |
| Visual add/remove | **Stable** | Cards, charts, shapes/buttons, images (auto-embed local files via `sourcePath`), textboxes, slicers — with full data bindings, bounds clamping, and ResourcePackageItem image references |
| Visual formatting | **Stable** | `pbix_format_visual` — human-readable API for titles, backgrounds, borders, drop shadows, padding, spacing, data labels, legend, axis, colors, table alternating row colors (backColorPrimary/Secondary, fontColorPrimary/Secondary), grid line colors, and 25+ more categories. Per-series/category dataColors with auto-generated selectors. Ground truth validated against PBI Desktop |
| Color extraction & recolor | **Stable** | `pbix_extract_colors` scans themes + all visuals. `pbix_recolor` replaces hex + ThemeDataColor refs, auto-extends palette, injects per-series/category chart colors, generates themed table rows, strips borders and pie/donut backgrounds, hides card titles (shows categoryLabels), fixes text contrast (WCAG 2.0) including theme foreground, chart axis/legend/labels, table rows, and card calloutValue |
| Visual property editing | **Stable** | Dot-path and full JSON |
| DAX measure CRUD | **Stable** | Add, modify, remove via binary splice (PBI Desktop files) or full builder rebuild. Sequential adds supported with automatic MAXID tracking |
| DAX evaluation (156 functions) | **Stable API** | Best-effort semantic parity — stable API, practical evaluation for common DAX patterns; see accuracy notes below |
| Metadata SQL read/write | **Stable** | Full SQLite access to tables, columns, relationships |
| Default slicer filter extraction | **Stable** | Legacy Layout JSON and PBIR format |
| Table data read | **Stable** | Native VertiPaq decoder — all materialized tables (no external dependencies) |
| Calculated table evaluation | **Stable** | DATATABLE, GENERATESERIES, CALENDAR, field parameters |
| XPress9 decompress/recompress | **Stable** | Byte-exact round-trip verified |
| ABF archive manipulation | **Stable** | List, extract, replace internal files |
| DataMashup (M code) editing | **Stable** | Read/write Power Query expressions |
| File save/repack | **Stable** | Auto-backup on overwrite, SecurityBindings auto-removed, optional MSIP sensitivity label stripping |
| Calculated column evaluation | **Beta** | Per-row DAX expression evaluation; tested with synthetic data |
| Password extraction | **Beta** | Regex scan of DAX measures for embedded passwords |
| Row-Level Security (RLS) | **Stable** | Read, write, and evaluate RLS roles. `set_rls_role` uses binary splice — roles persist across save/reopen. MAXID-based ID allocation. Verified on PBI Desktop files |
| Bookmark creation | **Beta** | Create/remove bookmarks with page targeting and visual visibility state |
| Field Parameters | **Stable** | Create field parameter tables via `pbix_datamodel_add_field_parameter` — uses full DataModel rebuild for VertiPaq storage |
| Calculation Groups | **Stable** | Create calculation groups via `pbix_datamodel_add_calculation_group` — table with CalculationItem DAX expressions, Partition Type=7, DiscourageImplicitMeasures enforced |
| TMDL Export | **Stable** | Export data model as Git-friendly TMDL text files via `pbix_export_tmdl`. Validated with Adventure Works DW 2020 — correct partition types, CrossFilteringBehavior, model properties, shared expressions |
| PBIP Export | **Stable** | Convert PBIX to PBIP (Power BI Project) folder structure via `pbix_export_pbip` — full TMDL semantic model + report layout + static resources, ready for Git |
| Perspectives | **Stable** | Create/list/remove perspectives via `pbix_add_perspective`, `pbix_get_perspectives`, `pbix_remove_perspective` |
| User Hierarchies | **Stable** | Create/list/remove drill-down hierarchies via `pbix_add_hierarchy`, `pbix_get_hierarchies`, `pbix_remove_hierarchy`. Works with builder-created and PBI Desktop files |
| Cultures & Translations | **Stable** | Add cultures, translate table/column/measure names via `pbix_add_culture`, `pbix_add_translations`, `pbix_get_cultures`, `pbix_remove_culture` |
| Partition Management | **Partial** | List/remove partitions via `pbix_get_partitions`, `pbix_remove_partition`. `pbix_add_partition` blocked for PBIX (needs PartitionStorage in VertiPaq), works for PBIP/TMDL export |
| Sensitivity Labels | **Stable** | Strip MSIP sensitivity labels via `pbix_save(strip_sensitivity_label=True)` |
| Custom Visuals | **Beta** | Import .pbiviz packages via `pbix_add_custom_visual`, place with `pbix_add_visual` |
| Incremental Refresh | **Stable** | `pbix_set_incremental_refresh` / `pbix_get_incremental_refresh` — configure archive/refresh windows with change detection. Requires data source (source_csv/source_db); embedded-only files cannot use incremental refresh (same as PBI Desktop) |
| Report diff (`pbix_diff`) | **Stable** | Compare two PBIX files — tables, columns, measures, relationships, pages/visuals, data sources, theme colors. Shows added/removed/changed |
| Report documentation (`pbix_document`) | **Stable** | Auto-generate full report documentation (markdown + .docx) — tables, columns, measures, relationships, data sources, pages/visuals, RLS roles, theme colors |
| Performance analysis (`pbix_performance`) | **Stable** | Flags oversized tables, empty tables, wide schemas, high-cardinality strings, complex measures, inactive/bidirectional relationships, orphaned tables |
| Diagnostic tool (`pbix_doctor`) | **Stable** | 17-point comprehensive diagnostic — data sources, storage modes, columns, relationships, measures, RLS, VertiPaq row counts, table/storage consistency, referential integrity, Expression/DataMashup consistency, MAXID |

## Known Limitations

- **DAX engine is best-effort** — designed for practical evaluation, not semantic parity with Analysis Services. Unsupported functions return `None` with status `"unsupported"`, circular references raise `DAXEvaluationError`. See [docs/supported-dax.md](docs/supported-dax.md) for full details.
- **PBIR format** — PBI Desktop (March 2026) has rendering bugs with PBIR decomposed format. PBIP export uses legacy report format (version 1.0) which works reliably.
- **1 out of 204 tested measures** returns BLANK (requires per-employee RANKX visual row context)
- **Performance** — tables >100K rows trigger a warning; the DAX engine operates on in-memory Python data
- **Opening existing DirectQuery files** — layout, measures, and metadata editing work; DAX evaluation and table reads return clear errors since data lives in the remote source (this is inherent to DirectQuery — the data isn't in the file)
- **Creating DirectQuery files** — fully working with SQL Server (LocalDB), PostgreSQL 16, and MySQL 9.6 (via MariaDB adapter); requires a running database server and initial data snapshot
- **CryptKey.bin** — the 144-byte RSA key BLOB cannot be generated without Microsoft's crypto infrastructure (`rskeymgmt`). A known-valid GUID-independent constant is used.
- **Embedded VertiPaq data** — verified working with 11 tables, 72 columns, 13 relationships, 121K+ rows (Adventure Works DW 2020) and 6 tables, 36 columns, 5 relationships, 25 rows, 3 pages, 14 visuals (Northwind showcase)
- **RLE encoding** — disabled in the VertiPaq encoder (pure bitpack used). Slightly less space-efficient but correct
- **Adding partitions to PBIX** — `pbix_add_partition` is blocked for PBIX files (needs PartitionStorage in VertiPaq). Works for PBIP/TMDL export. Reading and removing existing partitions works.
- **Full DataModel rebuild** — `set_table_data`, `update_table_rows`, `add/remove_relationship`, `remove_table` trigger a full DataModel rebuild via the builder pipeline. Most other tools (`add_measure`, `modify_measure`, `modify_column`, `set_rls_role`, `add_perspective`, `add_culture`, `add_translations`, `update_data_source`, etc.) use a lightweight metadata-only path.

## Tools (101)

### Create & File Management (5)
`pbix_create` · `pbix_open` · `pbix_save` · `pbix_close` · `pbix_list_open`

### Report Layout & Visuals (21)
`pbix_add_visual` · `pbix_remove_visual` · `pbix_format_visual` · `pbix_get_pages` · `pbix_add_page` · `pbix_remove_page` · `pbix_get_page_visuals` · `pbix_get_visual_detail` · `pbix_get_visual_positions` · `pbix_set_visual_property` · `pbix_update_visual_json` · `pbix_get_layout_raw` · `pbix_set_layout_raw` · `pbix_get_filters` · `pbix_set_filters` · `pbix_get_default_filters` · `pbix_get_settings` · `pbix_set_settings` · `pbix_get_bookmarks` · `pbix_add_bookmark` · `pbix_remove_bookmark`

### DAX Engine (4)
`pbix_evaluate_dax` · `pbix_evaluate_dax_per_dimension` · `pbix_evaluate_calculated_columns` · `pbix_clear_dax_cache`

### DataModel Read (16)
`pbix

…

## Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [d0nk3yhm](https://github.com/d0nk3yhm)
- **Source:** [d0nk3yhm/pbix-mcp](https://github.com/d0nk3yhm/pbix-mcp)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/mcp-d0nk3yhm-pbix-mcp
- Seller: https://agentstack.voostack.com/s/d0nk3yhm
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.