OpenShift Dependency Tree

Maps Go module dependencies, openshift/api usage, and repo metadata across ~250 OpenShift repositories. Answers questions like "which repos need to change for feature X?" using keyword-scored impact analysis.

Prerequisites

• Python 3.10+
• GitHub CLI (gh) authenticated with access to the openshift org
• mcp Python package (for the MCP server only): pip install mcp

Pipeline

Run the scripts in order. Each step caches its output so re-runs are fast.

# 1. Fetch go.mod files from all openshift/* repos → .cache/deps.json
python fetch_deps.py

# 2. Build the dependency graph from cached go.mod data
python build_graph.py --json > graph.json

# 3. Map which openshift/api packages and CRD kinds each repo uses
python analyze_api_usage.py --json > api_usage.json

# 4. Fetch GitHub metadata and auto-classify repos by platform/category
python fetch_repo_metadata.py

# 5. Score repos by relevance to a feature description
python feature_impact.py --feature "storage encryption at rest"

Scripts

fetch_deps.py

Fetches go.mod files from every openshift/* repo via the GitHub API and caches them under .cache/. This populates the raw dependency data that all other scripts build on.

build_graph.py

Builds the inter-repo dependency graph from cached go.mod data. Supports multiple output formats.

python build_graph.py                          # summary stats
python build_graph.py --dot                    # DOT output for Graphviz
python build_graph.py --focus api client-go    # who depends on these repos?
python build_graph.py --json > graph.json      # JSON for other tools

analyze_api_usage.py

Maps which openshift/api packages (API groups/versions) and CRD kinds each repo imports. Uses GitHub code search with caching.

python analyze_api_usage.py                    # build cache and print summary
python analyze_api_usage.py --repo cluster-etcd-operator
python analyze_api_usage.py --package config/v1
python analyze_api_usage.py --top-packages     # most-imported packages
python analyze_api_usage.py --json > api_usage.json

fetch_repo_metadata.py

Fetches GitHub metadata (description, topics, fork status, stars) for every repo and auto-classifies them by platform and category using keyword rules.

Platforms: aws, azure, gcp, vsphere, baremetal, openstack, ibm, nutanix, ovirt, kubevirt

Classifications: storage, networking, monitoring, security, installation, machine-management, cloud-compute, operator-framework, image-registry, update, scaling, console, etcd, scheduling, build, ci-testing, backup-restore, node, managed-services, ai-ml, cluster-lifecycle, credentials, hardware, edge, cli, policy, api-sdk

python fetch_repo_metadata.py                 # fetch all repos
python fetch_repo_metadata.py --refresh       # bust cache and re-fetch
python fetch_repo_metadata.py --repo installer # inspect one repo

Supports a repo_metadata_overrides.json file for manual corrections to auto-classification.

feature_impact.py

Scores repos by relevance to a feature description. Combines keyword matching across repo names, descriptions, topics, dependency graphs, and API package usage.

python feature_impact.py --feature "storage encryption at rest"
python feature_impact.py --feature "dual stack IPv6" --platform aws azure
python feature_impact.py --feature "etcd backup" --classification etcd --top 20
python feature_impact.py --feature "ingress TLS" --json
python feature_impact.py --feature "NVMe support" --output impact.json --min-score 20

view.html

Browser-based visualization. Open directly in a browser — no server required. Loads graph.json, api_usage.json, and repo_metadata.json to display an interactive dependency map with repo detail panels, platform/classification chips, and a Feature Impact tab.

MCP Server

mcp_server.py exposes the analysis as a Model Context Protocol server, allowing LLMs to query the dependency tree directly.

Tools

Tool	Description
feature_impact_tool	Score repos by relevance to a feature description
get_repo_info	Full detail for one repo (metadata, deps, API usage)
list_repos	List repos with platform/classification/fork filters
get_repo_dependencies	Forward and reverse dependency relationships
get_repo_api_usage	openshift/api packages and CRD kinds used by a repo
search_repos	Substring search across repo names, descriptions, topics

Resources

Resource	Description
openshift://filters	Valid platform and classification filter values
openshift://data-freshness	Data file existence, freshness, and record counts

Setup

Register with Claude Code:

claude mcp add openshift-dep-tree python3 /path/to/mcp_server.py

Or add to .claude/settings.local.json:

{
  "mcpServers": {
    "openshift-dep-tree": {
      "command": "python3",
      "args": ["/path/to/mcp_server.py"]
    }
  }
}

Container

Build and run the MCP server in a UBI 9 container. Data files are mounted at runtime so you can refresh them without rebuilding.

# Build
podman build -t openshift-dep-tree -f Containerfile .

# Run (mount project root as the data directory)
podman run -i --rm -v "$(pwd):/opt/app-root/src/data:Z" openshift-dep-tree

Register the container with Claude Code:

claude mcp add openshift-dep-tree \
  podman run -i --rm -v /path/to/openshift-dep-tree:/opt/app-root/src/data:Z openshift-dep-tree

The MCP_DATA_DIR environment variable controls where the server looks for data files (graph.json, api_usage.json, repo_metadata.json, .cache/). It defaults to the script directory when running outside a container.

Testing

Use the MCP Inspector for interactive testing:

npx @modelcontextprotocol/inspector python3 mcp_server.py

Data Files

File	Generated By	Description
.cache/deps.json	fetch_deps.py	Raw go.mod dependency data
graph.json	build_graph.py	Inter-repo dependency graph
api_usage.json	analyze_api_usage.py	openshift/api package and CRD kind usage
repo_metadata.json	fetch_repo_metadata.py	GitHub metadata with platform/classification tags