Twilize MCP Server

twilize

Tableau Workbook (.twb/.twbx) generation toolkit for reproducible dashboards and workbook engineering Programmatically create Tableau workbooks with stable analytical primitives, dashboard composition, and built-in structural validation.

Install in one click

If you just want to use twilize from an MCP client and skip the Python setup entirely, install it through the Smithery MCP registry:

• Install twilize on Smithery — one-click install for Cursor, Claude Desktop, VS Code, Claude Code, Windsurf, and the Smithery CLI.

Smithery writes the uvx twilize launch command directly into your MCP client's config. You need uv installed locally (Smithery will prompt you if it's missing) — nothing else.

If you'd rather wire the server up by hand, or you want the full Python library API, keep reading.

Overview

twilize is a Model Context Protocol (MCP) server and Python toolkit for generating Tableau Desktop workbook files (.twb / .twbx) from code or AI-driven tool calls.

It is designed as a workbook engineering layer, not as a conversational data exploration agent. The goal is to make workbook generation reproducible, inspectable, and safe to automate in local workflows, scripts, and CI.

The default workflow is:

1. Start from a known template (.twb or .twbx) or the built-in zero-config template
2. Add calculated fields and parameters
3. Build worksheets from stable chart primitives
4. Assemble dashboards and interactions
5. Save and validate a .twb or .twbx that opens in Tableau Desktop

                            Interfaces
  ┌───────────────────────────────────────────────────────────────┐
  │  ┌──────────────────────────┐  ┌───────────────────────────┐  │
  │  │        MCP Server        │  │      Python Library       │  │
  │  │  tools_workbook          │  │  from twilize.twb_editor    │  │
  │  │  tools_layout            │  │  import TWBEditor         │  │
  │  │  tools_migration         │  │                           │  │
  │  │  tools_support           │  │  editor.add_...()         │  │
  │  │                          │  │  editor.configure_...()   │  │
  │  │  (Claude / Cursor /      │  │  editor.save(...)         │  │
  │  │   VSCode / Claude Code)  │  │                           │  │
  │  └─────────────┬────────────┘  └──────────────┬────────────┘  │
  │                └──────────────┬────────────────┘               │
  └─────────────────────────────  ┼  ─────────────────────────────┘
                                  ▼
  ┌───────────────────────────────────────────────────────────────┐
  │                          TWBEditor                            │
  │       ParametersMixin  ·  ConnectionsMixin                    │
  │       ChartsMixin      ·  DashboardsMixin                     │
  └──────────┬──────────────────┬──────────────────┬─────────────┘
             ▼                  ▼                  ▼
  ┌──────────────────┐  ┌──────────────┐  ┌──────────────────────┐
  │  Chart Builders  │  │  Dashboard   │  │  Analysis &          │
  │                  │  │  System      │  │  Migration           │
  │  Basic  DualAxis │  │              │  │                      │
  │  Pie    Text     │  │  layouts     │  │  migration.py        │
  │  Map    Recipes  │  │  actions     │  │  twb_analyzer.py     │
  │                  │  │  dependencies│  │  capability_registry │
  └────────┬─────────┘  └──────┬───────┘  └──────────┬───────────┘
           └───────────────────┼──────────────────────┘
                               ▼
  ┌───────────────────────────────────────────────────────────────┐
  │                     XML Engine  (lxml)                        │
  │    template.twb/.twbx  →  patch  →  validate  →  save        │
  └───────────────────────────────┬───────────────────────────────┘
                                  ▼
                      output.twb  /  output.twbx

Installation

pip install twilize

To run the bundled Hyper-backed example that inspects .hyper files and resolves the physical Orders_* table automatically, install the optional example dependency as well:

pip install "twilize[examples]"

Requirements

• Python >= 3.10
• lxml >= 5.0
• uv
• mcp >= 1.0

Quick Start

As MCP Server

To allow an MCP client to build Tableau workbooks automatically, add twilize to that client's MCP configuration.

The launch command is the same across clients:

uvx twilize

Each client stores this command in a different configuration format. Use the matching example below.

Claude Desktop

Open ~/Library/Application Support/Claude/claude_desktop_config.json on macOS or %APPDATA%\Claude\claude_desktop_config.json on Windows and add:

{
  "mcpServers": {
    "twilize": {
      "command": "uvx",
      "args": ["twilize"]
    }
  }
}

Cursor IDE

1. Open Cursor Settings -> Features -> MCP
2. Click Add New MCP Server
3. Set Type to command
4. Set Name to twilize
5. Set Command to uvx twilize

Claude Code

claude mcp add twilize -- uvx twilize

VSCode

Open the workspace .vscode/mcp.json file or your user-profile mcp.json file and add:

{
  "servers": {
    "twilize": {
      "command": "uvx",
      "args": ["twilize"]
    }
  }
}

In VSCode, you can open these files from the Command Palette with MCP: Open Workspace Folder Configuration or MCP: Open User Configuration. You can also use MCP: Add Server and enter the same uvx twilize command through the guided flow.

Windsurf

Open ~/.codeium/windsurf/mcp_config.json and add:

{
  "mcpServers": {
    "twilize": {
      "command": "uvx",
      "args": ["twilize"]
    }
  }
}

Hosted (HTTP) server — no local Python required

If you do not want to install Python or uv locally, point your MCP client at the hosted twilize server. This loses local file access (the server cannot read CSVs or write .twbx files into your ~/Downloads), so it is best for read-only workflows like template recommendation, capability inspection, and remote workbook validation.

{
  "mcpServers": {
    "twilize": {
      "type": "streamable-http",
      "url": "https://web-production-dcbcb.up.railway.app/mcp"
    }
  }
}

As Python Library

Use TWBEditor(...) to start from a template and rebuild workbook content. Use TWBEditor.open_existing(...) when you want to keep existing worksheets and dashboards and reconfigure a sheet in place.

from twilize.twb_editor import TWBEditor

editor = TWBEditor("")  # "" uses the built-in Superstore template
editor.clear_worksheets()
editor.add_calculated_field("Profit Ratio", "SUM([Profit])/SUM([Sales])")

editor.add_worksheet("Sales by Category")
editor.configure_chart(
    worksheet_name="Sales by Category",
    mark_type="Bar",
    rows=["Category"],
    columns=["SUM(Sales)"],
)

editor.add_worksheet("Segment Pie")
editor.configure_chart(
    worksheet_name="Segment Pie",
    mark_type="Pie",
    color="Segment",
    wedge_size="SUM(Sales)",
)

editor.add_dashboard(
    dashboard_name="Overview",
    worksheet_names=["Sales by Category", "Segment Pie"],
    layout="horizontal",
)

editor.save("output/my_workbook.twb")

Working with Packaged Workbooks (.twbx)

.twbx files are ZIP archives that bundle the workbook XML together with data extracts (.hyper) and image assets. twilize reads and writes them transparently:

from twilize.twb_editor import TWBEditor

# Open a packaged workbook — extracts and images are preserved automatically
editor = TWBEditor.open_existing("templates/dashboard/MyDashboard.twbx")

# Make changes as usual
editor.add_calculated_field("Profit Ratio", "SUM([Profit])/SUM([Sales])")

# Save as .twbx — re-bundles the updated .twb with the original extracts/images
editor.save("output/MyDashboard_v2.twbx")

# Or extract just the XML when the packaged format isn't needed
editor.save("output/MyDashboard_v2.twb")

A plain .twb can also be packaged:

editor = TWBEditor("templates/twb/superstore.twb")
# ...
editor.save("output/superstore.twbx")  # produces a single-entry ZIP with the .twb inside

MCP Tools

Capability Model

Core primitives

These are the stable building blocks the project should continue to promise:

• Bar
• Line
• Area
• Pie
• Map
• Text / KPI cards
• Parameters and calculated fields
• Basic dashboard composition

Advanced patterns

These are supported, but they are higher-level compositions or interaction features rather than the default surface area:

• Scatterplot
• Heatmap
• Tree Map
• Bubble Chart
• Dual Axis — mark_color_1/2, color_map_1, reverse_axis_1, hide_zeroline, synchronized
• Table Calculations — RANK_DENSE, RUNNING_SUM, WINDOW_SUM via add_calculated_field(table_calc="Rows")
• KPI Difference badges — MIN(1) dummy axis + axis_fixed_range + color_map + customized_label
• Donut (via extra_axes) — multi-pane Pie + white circle using configure_dual_axis(extra_axes=[...]); supports color_map for :Measure Names palette
• Rich-text labels — configure_chart(label_runs=[...]) for multi-style KPI cards and dynamic titles with inline field values
• Advanced worksheet styling — configure_worksheet_style supports pane-level cell/datalabel/mark styles, per-field label/cell/header formats, axis tick control, tooltip disabling, and all Tableau visual noise suppressions
• Row dimension header suppression — configure_worksheet_style(hide_row_label="FieldName")
• Filter zones, parameter controls, color legends
• Dashboard filter and highlight actions
• Declarative JSON layout workflows
• Dashboard zone title control via show_title: false in layout dicts

Recipes and showcase patterns

These can be generated today, but they should be treated as recipes or examples rather than first-class promises:

• Donut
• Lollipop
• Bullet
• Bump
• Butterfly
• Calendar

Recipe charts are intentionally exposed through a single configure_chart_recipe tool so the public MCP surface does not grow one tool at a time for every showcase pattern.

This distinction matters because twilize is not trying to become a chart zoo or compete with Tableau's own conversational analysis tooling. The project is strongest when it provides a reliable, automatable workbook generation layer.

Capability-first workflow

When you are not sure whether something belongs in the stable SDK surface:

1. Use list_capabilities to inspect the declared boundary
2. Use describe_capability to check a specific chart, encoding, or feature
3. Use analyze_twb or diff_template_gap before chasing a showcase template

This keeps new feature work aligned with the project's real product boundary instead of with whatever happens to appear in a sample workbook.

Built-in Validation

Structural validation

save() automatically validates the TWB XML structure before writing:

• Fatal errors such as missing <workbook> or <datasources> raise TWBValidationError
• Warnings such as missing <view> or <panes> are logged but do not block saving
• Validation can be disabled with editor.save("output.twb", validate=False) or editor.save("output.twbx", validate=False)

XSD schema validation

TWBEditor.validate_schema() checks the workbook against the official Tableau TWB XSD schema (2026.1), vendored at vendor/tableau-document-schemas/:

result = editor.validate_schema()
print(result.to_text())
# PASS  Workbook is valid against Tableau TWB XSD schema (2026.1)
# — or —
# FAIL  Schema validation failed (2 error(s)):
#   * Element 'workbook': Missing child element(s)...

result.valid          # bool
result.errors         # list[str] — lxml error messages
result.schema_available  # False if the vendor submodule is not checked out

The same check is available as an MCP tool:

validate_workbook()                       # validate current open workbook in memory
validate_workbook(file_path="out.twb")    # validate a file on disk (.twb or .twbx)

XSD errors are informational — Tableau itself generates workbooks that occasionally deviate from the schema — but recurring errors signal structural problems worth fixing.

Dashboard Layouts

Custom layouts can be built programmatically using a nested layout dictionary or via generate_layout_json for MCP workflows.

Hyper-backed Example

The examples/hyper_and_new_charts.py example uses the Sample - EU Superstore.hyper extract bundled directly in the package (src/twilize/references/) and resolves the physical Orders_* table via Tableau Hyper API before switching the workbook connection. No repository clone is needed — install with pip install "twilize[examples]" and run directly.

Workbook Migration

twilize includes a migration subsystem for switching an existing .twb to a new datasource — for example, repointing a workbook built on one Excel file to a different Excel with a different schema, or migrating between language variants of the same dataset.

How it works

Migration is a multi-step workflow. Each step is available as both an MCP tool and a Python function:

1. inspect_target_schema   →  Scan the target Excel and list its columns
2. profile_twb_for_migration  →  Inventory which fields the workbook uses
3. propose_field_mapping   →  Match source fields to target columns (fuzzy)
4. preview_twb_migration   →  Dry-run: show what would change, blockers/warnings
5. apply_twb_migration     →  Write the migrated .twb + JSON reports

migrate_twb_guided is a convenience wrapper that runs steps 2–5 in sequence and pauses automatically when only low-confidence field matches remain, returning a warning_review_bundle for human review before proceeding.

Python example

from twilize.migration import migrate_twb_guided_json
import json

# One-call guided migration
result = migrate_twb_guided_json(
    file_path="templates/SalesDashboard.twb",
    target_source="data/new_data_source.xlsx",
    output_path="output/SalesDashboard_migrated.twb",
)
bundle = json.loads(result)

if bundle["status"] == "warning_review_required":
    # Inspect low-confidence matches and confirm or override them
    print(bundle["warning_review_bundle"])
    # Re-run with confirmed mappings
    result = migrate_twb_guided_json(
        file_path="templates/SalesDashboard.twb",
        target_source="data/new_data_source.xlsx",
        output_path="output/SalesDashboard_migrated.twb",
        mapping_overrides={"Old Field Name": "New Column Name"},
    )

MCP tool example

When using twilize as an MCP server, an AI agent can run the full workflow:

inspect_target_schema(target_source="data/new_data_source.xlsx")
→ returns column list and data types

migrate_twb_guided(
    file_path="templates/SalesDashboard.twb",
    target_source="data/new_data_source.xlsx",
    output_path="output/SalesDashboard_migrated.twb"
)
→ returns status: "applied" or "warning_review_required"

Output files

A completed migration writes three files:

Scope parameter

scope="workbook" migrates all worksheets. Pass a worksheet name to limit migration to a single sheet.

Self-contained example

examples/migrate_workflow/ contains a template .twb, the original Superstore Excel, a target Chinese-locale Superstore Excel, and a runnable script:

python examples/migrate_workflow/test_migration_workflow.py

Project Structure

twilize/
|-- src/twilize/
|   |-- __init__.py
|   |-- capability_registry.py
|   |-- config.py
|   |-- charts/
|   |-- connections.py
|   |-- dashboard_actions.py
|   |-- dashboard_dependencies.py
|   |-- dashboard_layouts.py
|   |-- dashboards.py
|   |-- field_registry.py
|   |-- layout.py
|   |-- layout_model.py
|   |-- layout_rendering.py
|   |-- mcp/
|   |-- parameters.py
|   |-- twb_analyzer.py
|   |-- twb_editor.py
|   |-- validator.py
|   `-- server.py
|-- tests/
|-- examples/
|-- docs/
|-- pyproject.toml
`-- README.md

Development

# Install in editable mode
pip install -e .

# Run test suite
pytest --basetemp=output/pytest_tmp

# Run the mixed showcase example
python examples/scripts/demo_all_supported_charts.py

# Run the advanced Hyper-backed example
python examples/scripts/demo_hyper_and_new_charts.py

# Run the guided migration example
python examples/migrate_workflow/test_migration_workflow.py

# Start MCP server
twilize

MCP Server Manifest

twilize ships with a full MCP server manifest (mcp-server.json) — the MCP equivalent of a Tableau Extension .trex file. It declares:

Additionally, the manifest declares all 40 tools with categories, 6 resources, and pre-built client configurations for Claude Desktop, Cursor, VS Code, and Claude Code.

For the complete tool reference with parameter schemas, see docs/MCP_SERVER.md.

Publishing

PyPI

pip install build twine
python -m build
twine upload dist/*

Smithery MCP Registry

The smithery.yaml file is ready for submission to Smithery.

MCP Server Registry

The server.json file follows the official MCP Registry schema and is ready to publish via the mcp-publisher CLI.

# 1. Install the publisher CLI (macOS / Linux)
curl -L "https://github.com/modelcontextprotocol/registry/releases/latest/download/mcp-publisher_$(uname -s | tr '[:upper:]' '[:lower:]')_$(uname -m | sed 's/x86_64/amd64/;s/aarch64/arm64/').tar.gz" \
    | tar xz mcp-publisher && sudo mv mcp-publisher /usr/local/bin/

# 2. Authenticate with the GitHub account that owns the namespace
mcp-publisher login github

# 3. Publish the contents of server.json (run from repo root)
mcp-publisher publish

# 4. Confirm
curl "https://registry.modelcontextprotocol.io/v0.1/servers?search=io.github.subhatta123/twilize"

The legacy mcp-server.json (older schema used by the modelcontextprotocol/servers README) is also kept for backwards compatibility with clients that scrape that list.

Hosted (HTTP) MCP server

The repository ships two Railway configs:

• railway.json — builds the browser dashboard extension (FastAPI + Vite frontend at /).
• railway.mcp.json — builds the HTTP MCP server from Dockerfile.mcp and exposes it on /mcp for clients that connect over streamable-http.

To deploy the MCP server as a second Railway service:

1. In your Railway project, + New → GitHub Repo → subhatta123/twilize.
2. Open the new service's Settings → Config-as-code Path and set it to railway.mcp.json.
3. Networking → Generate Domain to mint a public URL.
4. Verify with curl https://<your-domain>/mcp — a JSON-RPC text/event-stream error confirms the server is responding.

Attribution

Twilize builds upon and was inspired by cwtwb, a Python-based open-source tool for programmatically generating and automating the migration of Tableau Workbooks (.twb files), originally authored by @imgwho.

cwtwb is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0). In accordance with that license, Twilize is also distributed under AGPL-3.0.

License

AGPL-3.0-or-later