pbirb-mcp

An MCP server for editing Power BI Report Builder paginated reports (.rdl) through Claude (Desktop, CLI, or any MCP client). Forty-plus tools cover the gaps that otherwise force hand-written XML: dataset filters, headers and footers, body composition, groupings, sorting, row visibility, conditional expressions, styling, page setup, advanced parameters, and embedded images.

The server speaks JSON-RPC 2.0 over stdio. It opens an .rdl from disk, mutates it in place via lxml, validates structure, and writes atomically — a failed save never leaves a half-written report or scrubs the original.

Stability

Pre-1.0. The tool surface — tool names, inputSchema, output shapes, error semantics — is the contract. While on 0.x, MINOR releases may include a small breaking change with a migration note in CHANGELOG.md; after v1.0, breaking changes require MAJOR. See CONTRIBUTING.md for the full bump rules adapted from SemVer for an MCP tool surface.

Pin to a MINOR while on 0.x (e.g. pbirb-mcp~=0.1) if your prompts depend on specific tool names or schemas.

Related MCP server: Power BI MCP for Claude

Quick start

1. Install

The simplest path is uv + PyPI — no clone, no venv, no install step:

uvx pbirb-mcp

uvx fetches the package into a throwaway environment, runs the pbirb-mcp console script, and exits. The MCP server speaks JSON-RPC over stdio, so any MCP client (Claude Desktop, Claude Code, etc.) can spawn it directly.

For local development against this codebase instead:

git clone https://github.com/mafaq229/pbirb-mcp
cd pbirb-mcp
uv venv .venv
uv pip install --python .venv/bin/python -e ".[dev]"

Verify the binary works:

printf '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}\n' \
  | .venv/bin/pbirb-mcp

You should see a single JSON-RPC response with protocolVersion, capabilities.tools, and serverInfo.name = "pbirb-mcp".

2. Wire into Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "pbirb": {
      "command": "uvx",
      "args": ["pbirb-mcp"],
      "env": {
        "PBIRB_MCP_LOG_LEVEL": "INFO",
        "PBIRB_MCP_LOG_FILE": "/tmp/pbirb-mcp.log"
      }
    }
  }
}

Restart Claude Desktop. The hammer icon should show pbirb and the 40+ tools listed below.

For development against a local checkout, swap the args for ["--from", "/absolute/path/to/pbirb-mcp", "pbirb-mcp"] so uvx runs your working tree instead of the published version.

3. Wire into Claude Code

claude mcp add pbirb -- uvx pbirb-mcp

Or add to .mcp.json at the workspace root:

{
  "mcpServers": {
    "pbirb": {
      "command": "uvx",
      "args": ["pbirb-mcp"]
    }
  }
}

Or install it as a Claude Code plugin — one command instead of editing config by hand (it wires up the same uvx pbirb-mcp server for you):

/plugin marketplace add mafaq229/pbirb-mcp
/plugin install pbirb-mcp@pbirb

4. Smoke test

In any chat, ask the model:

Open ~/path/to/some-report.rdl and tell me what datasets and parameters it has.

The model should call describe_report, get_datasets, and get_parameters and summarise the result. If it doesn't, your client probably hasn't loaded the server — check the client's MCP debug log.

Working with a report

The recommended flow is read first, then edit by name. Every mutating tool addresses elements by their RDL Name attribute (Tablix name, Textbox name, ReportParameter name, DataSet name) — never by index. Indices break across multi-step edits.

A typical session:

You:    Bind the existing report at ~/Reports/sales.rdl to the Finance
        workspace's "FY26 Sales" dataset, change the body DAX to a date-
        ranged query, and add a header with a logo and a parameter line.

Claude: → describe_report(path=…)
        → set_datasource_connection(name="PBIDataset",
              workspace_url="Finance", dataset_name="FY26 Sales")
        → update_dataset_query(dataset_name="MainDataset",
              dax_body="EVALUATE FILTER('Sales',
                'Sales'[OrderDate] >= @DateFrom &&
                'Sales'[OrderDate] <= @DateTo)")
        → add_query_parameter(dataset_name="MainDataset",
              name="DateFrom",
              value_expression="=Parameters!DateFrom.Value")
        → add_query_parameter(dataset_name="MainDataset",
              name="DateTo",
              value_expression="=Parameters!DateTo.Value")
        → add_embedded_image(name="Logo",
              mime_type="image/png", image_path="~/brand/logo.png")
        → set_page_header(height="0.75in",
              print_on_first_page=True, print_on_last_page=True)
        → add_header_image(name="HeaderLogo",
              image_source="Embedded", value="Logo",
              top="0in", left="0in", width="1in", height="0.5in")
        → add_header_textbox(name="DateLine",
              text='="From: " & Parameters!DateFrom.Value & "  To: "
                    & Parameters!DateTo.Value',
              top="0.1in", left="1.2in", width="4in", height="0.25in")

Open the resulting .rdl in Report Builder; the change is visible in the designer and renders in Preview against the bound dataset.

Tool reference

42 tools, grouped by RDL concern. Every tool takes a path argument (absolute path to the .rdl); the rest of the schema is in tools/list output and visible to the LLM at registration time.

Read-only inventory

The "what's in this report?" tools. Always the first calls when planning a multi-step edit.

Datasource & dataset

Tablix

Page

Page header & footer

Same set of operations for each region, each accepts named items so follow-up edits don't drift on indices.

Body composition

Snippet templates

Single-call inserts of common report items, programmatically built and appended to the body.

Styling

Visibility

Parameters (advanced)

Cascading parameters

RDL has no <DependsOn> element — cascading is inferred from =Parameters!X.Value references in a lookup dataset's <QueryParameters>. To wire parameter B to depend on parameter A:

1. set_parameter_available_values(name="B", source="query", query_dataset="LookupB", ...)

2. add_query_parameter(dataset_name="LookupB", name="@A", value_expression="=Parameters!A.Value")

Report Builder figures out the dependency graph by parsing those expressions.

Embedded images

Reference an embedded image with add_*_image(image_source="Embedded", value="<image-name>").

Power BI specifics

XMLA connection strings

set_datasource_connection writes the canonical form:

Data Source=powerbi://api.powerbi.com/v1.0/myorg/<workspace>;Initial Catalog=<dataset>

workspace_url accepts a bare workspace name (Finance) or a full powerbi:// URL — the tool detects the latter and avoids double-prefixing. DataProvider is set to SQL (the Analysis Services provider id RDL uses for PBI XMLA, despite the misleading name).

DAX queries

DAX bodies are accepted verbatim — pbirb-mcp doesn't parse DAX, so the user (or Report Builder at preview time) is the source of truth for syntax. Empty bodies are rejected up front because Report Builder loads them but errors at preview, which is a worse signal than a clear ValueError here.

PBI paginated reports do not carry <CommandType> for DAX (unlike SSRS where you'd set CommandType=StoredProcedure); these tools never emit it.

Pre-commit hooks (contributors)

pre-commit is in the [dev] extras. After a fresh checkout:

uv pip install --python .venv/bin/python -e ".[dev]"
.venv/bin/pre-commit install                # one-time — installs the git hook
.venv/bin/pre-commit run --all-files        # one-time — clean any drift

After the hook is installed, every git commit runs ruff format + ruff check (with --fix) + the fast pytest suite. If lint or tests fail the commit is aborted; fix and re-stage before retrying.

To run individual hooks ad-hoc:

.venv/bin/pre-commit run ruff --all-files
.venv/bin/pre-commit run ruff-format --all-files
.venv/bin/pre-commit run pytest-fast --all-files

The full config lives in .pre-commit-config.yaml. Ruff settings (line length, rule selection, per-file ignores) live under [tool.ruff*] in pyproject.toml.

Report Builder install

Power BI Report Builder is a free Microsoft-distributed Windows app:

https://www.microsoft.com/en-us/download/details.aspx?id=105942

Open any .rdl produced by pbirb-mcp directly in Report Builder. The "opens cleanly with no upgrade prompt" check is the actual integration test — the unit tests verify schema correctness, but lxml will round-trip XML that Report Builder's deserialiser still rejects (we hit this twice in the chart-template work; both fixes are documented in the git history).

Architecture

pbirb-mcp/
├── pbirb_mcp_server.py         # Entry point (logging + main())
├── pbirb_mcp/
│   ├── server.py               # JSON-RPC stdio dispatch
│   ├── tools.py                # Tool registry — wires ops into the server
│   ├── core/
│   │   ├── document.py         # RDLDocument: open/save (lxml), atomic write
│   │   ├── xpath.py            # Namespace-aware XPath helpers
│   │   ├── ids.py              # Stable element addressing
│   │   └── schema.py           # Structural + opt-in XSD validation
│   ├── ops/
│   │   ├── reader.py           # describe / get_datasets / get_params / get_tablixes
│   │   ├── datasource.py       # PBI XMLA connection authoring
│   │   ├── dataset.py          # DAX body + query parameters
│   │   ├── tablix.py           # Filters, groupings, sort, visibility, height
│   │   ├── page.py             # Page setup + orientation
│   │   ├── header_footer.py    # Page header / footer authoring
│   │   ├── body.py             # Body textboxes / images / removal
│   │   ├── templates.py        # Chart + tablix snippet builders
│   │   ├── styling.py          # set_textbox_style + alternating row color
│   │   ├── visibility.py       # Element-level visibility
│   │   ├── parameters.py       # Available / default values, advanced flags
│   │   └── embedded_images.py  # Base64 image embedding
│   └── schemas/                # (Empty — drop the official RDL XSD here for opt-in XSD validation)
└── tests/
    ├── fixtures/
    │   └── pbi_paginated_minimal.rdl  # Hand-tuned to match Report Builder's emitted style
    └── test_*.py               # 270+ tests — every commit's tool plus round-trip invariants

Hard rules (enforced by tests)

• Tests first. Every commit writes failing tests, then makes them pass.

• lxml, not stdlib xml.etree. Round-trip fidelity is a feature, not polish. Report Builder reads what's on disk; formatting drift causes silent corruption.

• Stable IDs, never indices. Tools take tablix_name + group_name, not column_index: 2. Indices break across multi-step edits.

• Atomic save. RDLDocument.save_as writes to <path>.tmp then renames. A failure mid-write never leaves a half-written report.

• Round-trip byte-identity is enforced by tests/test_document.py's test_round_trip_byte_identical_to_fixture. A no-op open → save → reopen produces a byte-identical file.

RDL gotchas learned the hard way

• <?xml version="1.0" encoding="utf-8"?> uses double quotes, not lxml's default single quotes. Fixed in RDLDocument.save_as via a manual declaration.

• Self-closing tags use <Tag /> with a space, not <Tag/>. Fixed via a post-process regex.

• The rd: prefix (http://schemas.microsoft.com/SQLServer/reporting/reportdesigner) carries designer metadata Report Builder relies on. Don't strip it; preserve prefixes.

• Do not put MustUnderstand="df" on <Report> unless you also declare xmlns:df=....

• <ChartCategoryAxes> / <ChartValueAxes> hold <ChartAxis> children directly — there is no <ChartCategoryAxis> / <ChartValueAxis> wrapper.

• <ChartMember> requires a <Label> child, even an empty one.

• DAX queries live in <DataSet><Query><CommandText>EVALUATE ...</CommandText></Query></DataSet>. No <CommandType> element for DAX (unlike SSRS).

Logging

Two environment variables control the logger:

PBIRB_MCP_LOG_LEVEL=DEBUG PBIRB_MCP_LOG_FILE=/tmp/pbirb-mcp.log pbirb-mcp

Development

Running tests

.venv/bin/python -m pytest tests/ -v

The suite is fast (~0.3s for 270+ tests) so re-run on every change.

Smoke testing the live binary

printf '%s\n' \
  '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}' \
  '{"jsonrpc":"2.0","id":2,"method":"tools/list"}' \
  | .venv/bin/pbirb-mcp

For an end-to-end sanity check, drive an actual mutation against a copy of the bundled fixture:

SCRATCH=$(mktemp -d)/r.rdl
cp tests/fixtures/pbi_paginated_minimal.rdl "$SCRATCH"
.venv/bin/python -c "
from pbirb_mcp.ops.dataset import update_dataset_query
update_dataset_query(path='$SCRATCH', dataset_name='MainDataset',
    dax_body=\"EVALUATE TOPN(10, 'Sales')\")
print('Wrote', '$SCRATCH')
"

Open the resulting file in Power BI Report Builder. Manual verification that an .rdl opens cleanly is the actual integration test — the unit tests catch schema-level mistakes, but only Report Builder catches deserialiser nits.

Adding a new tool

1. Write tests first under tests/test_*.py.

2. Implement in the appropriate pbirb_mcp/ops/*.py module (or create a new one).

3. Register in pbirb_mcp/tools.py with a clear description and strict inputSchema.

4. Run the full suite and a JSON-RPC smoke against the fixture.

5. Open the modified RDL in Report Builder.

The commit-by-commit history shows the pattern in practice.

Releases

CHANGELOG.md tracks every release in Keep a Changelog format. Releases are also published as GitHub Releases and to PyPI.

Versions follow SemVer adapted for an MCP tool surface — see CONTRIBUTING.md § Versioning.

Contributing

PRs welcome. CONTRIBUTING.md covers dev setup, the hard rules (tests-first, lxml only, stable IDs, atomic save, byte-identity round-trip, smoke in Report Builder), the SemVer-for-MCP bump table, and the PR review checklist.

Bug reports and tool proposals: please use the issue templates. For security issues, see SECURITY.md. All participants are expected to follow the Code of Conduct.

Acknowledgments

The existing bethmaloney/rdl-mcp server pioneered the MCP-over-RDL pattern but is scoped to SSRS: column metadata, basic parameter management, stored-procedure swap. Power BI paginated reports use the same RDL 2016 schema as SSRS, but:

1. Data sources are Power BI XMLA endpoints, not SQL Server.

2. Queries are DAX, and the upstream tool exposes no body-edit (only stored-procedure name swap, useless here).

3. Report Builder is picky about XML round-tripping — formatting drift, namespace prefix loss, or unrecognised MustUnderstand attributes cause silent corruption or "this report needs to be upgraded" prompts.

pbirb-mcp is built around lxml so a no-op edit produces a byte-identical file, addresses every element by stable name (never index), and treats "opens cleanly in Report Builder" as the actual integration test.

mcp-name: io.github.mafaq229/pbirb-mcp