This website is built using Mintlify.

Create a pull request to make document changes, and request a review from @muhsinking. Changes will be deployed to production automatically after they are pushed to the main branch.

Install Mintlify:

npm i -g mintlify

Start a local development server on your browser:

mintlify dev

Most changes will be reflected live without having to restart the server.

• Install vale
• Lint a specific folder or file, run:

vale path/to/docs/
# or
vale path/to/*.md

Install blacken-docs.

python3 -m pip install blacken-docs

Run the formatter.

yarn format

git ls-files -z -- '*.md' | xargs -0 blacken-docs

# Set up virtual environment
python3 -m venv helpers/.venv
source helpers/.venv/bin/activate
pip install -r helpers/requirements.txt

# Run scripts
python3 helpers/gpu_types.py
python3 helpers/sls_cpu_types.py

The tests/TESTS.md file contains test definitions for validating documentation quality through AI agent testing. Tests simulate real user prompts - a coding agent must accomplish the goal using only the documentation as it currently exists.

• Claude Code with the Runpod MCP servers configured:

  # Add Runpod API MCP server
  claude mcp add runpod --scope user -e RUNPOD_API_KEY=your_key -- npx -y @runpod/mcp-server@latest
  
  # Add Runpod Docs MCP server
  claude mcp add runpod-docs --scope user --transport http https://docs.runpod.io/mcp

In Claude Code, use natural language:

Run the flash-quickstart test

Run all vLLM tests

To validate unpublished doc changes, use local docs mode:

Run the vllm-deploy test using local docs

Claude will:

1. Read the test from tests/TESTS.md
2. Attempt to accomplish the goal using only the docs
3. Clean up any resources created (prefixed with doc_test_)
4. Write a report to tests/reports/
5. Suggest documentation improvements

All tests are defined in tests/TESTS.md as a table

Add a row to the appropriate section in tests/TESTS.md with:

• ID: Unique test identifier
• Goal: One sentence describing what the user wants
• Cleanup: Resource types to delete (endpoints, pods, templates, network-volumes, or none)

Test reports are saved to tests/reports/ (gitignored) and include:

• What worked and what didn't
• Where the agent got stuck
• Specific documentation improvement suggestions