Build & Publish Pipeline

Relevant source files

The following files were used as context for generating this wiki page:

• pyproject.toml
• packages/google-workspace-mcp/pyproject.toml
• packages/markdowndeck/pyproject.toml
• packages/aws-s3-mcp/pyproject.toml
• README.md
• packages/google-workspace-mcp/README.md

The Build & Publish Pipeline for the Arclio MCP Tooling project is designed as a streamlined, monorepo-compatible system leveraging uv for package management and hatchling as the build backend. The pipeline facilitates the development, testing, and distribution of multiple Model Context Protocol (MCP) packages—including Google Workspace, AWS S3, and Weaviate integrations—ensuring consistent quality and standardized delivery across the ecosystem.

Sources: pyproject.toml:1-10, README.md:1-20

Pipeline Architecture

The pipeline follows a standardized structure for high-performance Python development within a monorepo. It utilizes workspace members to manage dependencies and shared resources across sub-packages.

Core Components

• Build System: Uses hatchling to generate distribution wheels and source distributions.
• Package Manager: Employs uv for fast dependency resolution and workspace orchestration.
• Task Runner: Managed through a root Makefile and poe (Poe the Poet) tasks defined in the configuration files.

The diagram shows the relationship between the root workspace configuration and the individual packages being built into distributable artifacts.
Sources: pyproject.toml:1-5, pyproject.toml:60-70, README.md:22-30

Build Configuration

Each package maintains its own pyproject.toml but is referenced by the root workspace. The build process targets Python 3.10+ and enforces metadata standards.

Workspace Members

The project defines the following packages as part of its buildable workspace:

Package Name	Path	Purpose
google-workspace-mcp	packages/google-workspace-mcp	Google API integrations
aws-s3-mcp	packages/aws-s3-mcp	S3 storage access
markdowndeck	packages/markdowndeck	Markdown to Slides conversion
weaviate-mcp	packages/weaviate-mcp	Vector database access

Sources: pyproject.toml:60-65, README.md:24-35

Distribution Packaging

Packages are built into wheels with a standardized directory structure. The hatchling backend is configured to find package source code within the src/ directory of each sub-package.

[tool.hatch.build.targets.wheel]
packages = ["packages/*"]

Sources: pyproject.toml:73-74, packages/google-workspace-mcp/pyproject.toml:34-35

Release & Publish Workflow

The publishing workflow is automated through poe tasks and Makefile commands. It involves cleaning artifacts, building the package, and uploading to a registry using uv.

Publication Logic

A specific sequence is followed for releases, particularly for the Google Workspace package:

1. Clean: Remove existing dist/ directories to ensure no stale artifacts are included.
2. Build: Invoke uv build within the package directory.
3. Publish: Upload artifacts using uv publish with a provided $PYPI_TOKEN.

The sequence diagram illustrates the automated multi-step release process for the Google Workspace MCP package.
Sources: pyproject.toml:76-90, packages/google-workspace-mcp/README.md:275-285

Release Commands

The following tasks are defined in the root pyproject.toml for managing the Google Workspace release lifecycle:

Task	Command	Description
build-gw	uv build	Compiles the package into distribution files
version-gw	python -c ...	Extracts the current version from pyproject.toml
publish-gw	uv publish	Uploads any version in dist/ to the registry
release-gw	clean, build, publish	Full release pipeline execution

Sources: pyproject.toml:77-88

Quality Assurance Pipeline

Before publishing, the codebase must pass a rigorous quality suite. This is orchestrated through a Makefile that handles environment setup and testing.

• Linting & Formatting: Uses Ruff for both linting and formatting across the entire monorepo.
• Testing: Orchestrated by pytest with specific markers for unit, integration, and end_to_end tests.
• Coverage: Configured to fail if coverage drops below 85% for core libraries like markdowndeck.

Sources: pyproject.toml:13-25, pyproject.toml:49-51, README.md:50-70

Dependency Management

The project uses uv workspace sources to allow packages to depend on each other during development (e.g., google-workspace-mcp depending on markdowndeck) without needing to fetch from an external registry.

[tool.uv.sources]
aws-s3-mcp = { workspace = true }
google-workspace-mcp = { workspace = true }
markdowndeck = { workspace = true }
weaviate-mcp = { workspace = true }

Sources: pyproject.toml:67-71, packages/google-workspace-mcp/pyproject.toml:23

Summary

The Build & Publish Pipeline ensures that Arclio's MCP tools are consistently tested and reliably distributed. By combining uv for workspace management, hatchling for standard-compliant builds, and automated poe tasks for registry publication, the project maintains a high velocity with minimal manual intervention. This system supports a "blank canvas first" architecture where individual components can be built and published independently or as a unified toolset.

Sources: README.md:38-45, packages/google-workspace-mcp/README.md:280-300