LLM-Friendly Documentation (llms.txt)

How to use the Konflux Operator llms.txt file with MCP servers so AI coding assistants can access the documentation.

The Konflux Operator documentation publishes an llms.txt file — a structured index that lets Large Language Models discover and consume the docs at inference time. The file is automatically generated from the Hugo content tree and stays in sync with documentation changes.

Published at: https://konflux-ci.dev/konflux-ci/llms.txt

What is llms.txt?

llms.txt is a lightweight specification for providing LLM-readable content on websites. It is a Markdown file with a title, description, and a list of links to documentation pages. AI coding assistants and MCP-enabled IDEs can read this file to find relevant documentation without scraping HTML.

Using with an MCP server (mcpdoc)

mcpdoc is an open-source MCP server purpose-built for serving llms.txt files. It exposes two tools to MCP host applications (Cursor, Windsurf, Claude Code/Desktop):

list_doc_sources — lists available documentation sources
fetch_docs — fetches content from URLs found in the llms.txt file

Konflux llms.txt URL

Use the following name and URL when configuring mcpdoc:

Name	URL
`Konflux Operator And Administrator Documentation`	`https://konflux-ci.dev/konflux-ci/llms.txt`

Quick start (Cursor example)

Open Cursor Settings > MCP to edit your ~/.cursor/mcp.json, then add:

{
  "mcpServers": {
    "konflux-operator-docs": {
      "command": "uvx",
      "args": [
        "--from",
        "mcpdoc",
        "mcpdoc",
        "--urls",
        "Konflux Operator And Administrator Documentation:https://konflux-ci.dev/konflux-ci/llms.txt",
        "--transport",
        "stdio"
      ]
    }
  }
}

For best results, add a rule to your Cursor User Rules (Settings > Rules):

For ANY question about Konflux, the Konflux Operator, or Konflux administration,
use the konflux-operator-docs MCP server:
1. Call list_doc_sources to see available llms.txt files
2. Call fetch_docs to read the llms.txt index
3. Identify URLs relevant to the question
4. Call fetch_docs on those URLs
5. Use the retrieved content to answer

Other IDEs and clients

mcpdoc supports Cursor, Windsurf, Claude Code, and Claude Desktop. For setup instructions for each client, see the mcpdoc Quickstart guide. The configuration is the same across clients — only the --urls value is Konflux-specific:

Konflux Operator And Administrator Documentation:https://konflux-ci.dev/konflux-ci/llms.txt

Testing with the MCP Inspector

Start mcpdoc in SSE mode:

uvx --from mcpdoc mcpdoc \
    --urls "Konflux Operator And Administrator Documentation:https://konflux-ci.dev/konflux-ci/llms.txt" \
    --transport sse \
    --port 8082 \
    --host localhost

Then launch the MCP Inspector:

npx @modelcontextprotocol/inspector

In the Inspector UI, set the transport to SSE, enter http://localhost:8082/sse, and click Connect. Use the Tools tab to call list_doc_sources and fetch_docs.

Testing with a local docs server

When developing documentation locally, point mcpdoc at the local Hugo server instead of the production URL:

make docs-serve

# In another terminal
uvx --from mcpdoc mcpdoc \
    --urls "Konflux Operator And Administrator Documentation:http://localhost:4000/konflux-ci/operator/llms.txt" \
    --allowed-domains localhost raw.githubusercontent.com \
    --transport sse \
    --port 8082 \
    --host localhost

When using a local `llms.txt` URL, you must pass `--allowed-domains` explicitly because mcpdoc only auto-allows the domain of the `llms.txt` URL itself. The documentation links inside `llms.txt` point to `raw.githubusercontent.com`, so that domain must be allowed as well.

How llms.txt is generated

The file is produced by hack/generate-llms-txt.sh, which scans the Hugo content tree and extracts title, description, and weight from each page’s YAML frontmatter. It runs automatically as part of make generate-docs, so any changes to the documentation are reflected in llms.txt on the next build.