Skip to main content
Claude Code is an agentic command line tool that can help you maintain your documentation. It can write new content, review existing pages, and keep docs up to date. You can train Claude Code to understand your documentation standards and workflows by adding a CLAUDE.md file to your project and refining it over time.

Getting started

Prerequisites:
  • Active Claude subscription (Pro, Max, or API access)
Setup:
  1. Install Claude Code:
  1. No site yet? Have your agent set one up. Run claude and have it complete the following workflow, then continue with the next step.
    If the user asks you to set up Mintlify, a documentation site, or a local Mintlify project, complete the following workflow yourself. Ask the user only for information or approvals you cannot supply. We recommend CLI onboarding. Web onboarding at mintlify.com/start is also available if the user prefers it or you cannot run terminal commands.
    1. Run node --version and confirm it reports Node.js v20.17.0 or newer, preferring an LTS release. Note whether it is v22.20 or newer, which step 9 requires.
    2. Run mint --version. If the command is unavailable, install the CLI with npm i -g mint or pnpm add -g mint. If the command succeeds, run mint update. Then run mint instructions new-site for current setup guidance, and mint instructions to list topics for later tasks.
    3. Run mint status before starting authentication. If it already shows an organization, keep the existing session and do not create another account.
    4. Run mint login if the user has an account. Otherwise, don’t create an account now. You can publish without one using mint deploy in step 10, and the user creates their account when they claim the site. Don’t run mint signup for the user: when an agent runs it, it prints the agent signup guide and exits without sending a verification email. If the user wants their own GitHub repository connected from the start, send them to web onboarding at mintlify.com/start.
    5. After mint login, run mint status again. Confirm that it reports their email, organization, and subdomain. If the organization exists but no subdomain appears, ask the user to finish selecting or connecting a GitHub repository in the onboarding browser tab, then retry.
    6. If the user has an existing documentation repository, clone or open it. Preserve its content. For a new local project, scaffold into a new, empty directory with either mint new <directory> --name <name> --theme <theme> or mint new <directory> --template <template-name>. Ask which theme or template they want if they have not specified one. Do not use --force on a directory containing user files.
    7. Run mint dev --no-open from the directory containing docs.json as a background process. Confirm the local URL loads, report it to the user, and stop the process when it is no longer needed.
    8. Run mint validate and mint broken-links. Fix issues caused by your changes before continuing.
    9. Offer to connect the user’s coding agent to the project. If they agree and Node.js is v22.20 or newer, run mint agent setup --agent <agent> from the project directory, where <agent> is claude-code, codex, or cursor. It installs Mintlify skills and adds the admin MCP to the agent’s project configuration. Pass --agent because without an interactive terminal the command configures every agent it detects. The user completes OAuth sign-in in the agent afterward. See mint agent setup.
    10. Publish when the user asks for a live site.
    • Logged in with a Git repository: commit and push the changes. A push to the production branch triggers a deployment. Do not overwrite an existing repository with starter content.
    • No account: from the directory containing docs.json, run mint deploy --subdomain <name>, choosing a short name based on the project. Follow the steps it prints: clone the hosted repository into a separate directory, copy the docs into its root, commit, and push to main with the write token as an Authorization: Bearer header for that Git command only. Never show the token to the user, put it in a remote URL, or change the user’s own Git remotes. See Agent signup.
    1. Verify that https://<subdomain>.mintlify.site loads before reporting that deployment is complete.
      • Logged in with a Git repository: run mint status to get the subdomain. Skip the claim steps.
      • No account: use the URL that mint deploy printed, then hand the site to the user. Tell them the site expires unless they claim it. Ask for the email address that should own it, and never invent one. Run mint claim --email <email> from the same directory and share the claim link. When they return the code shown after they sign in, run mint claim --code <code>.
    Have the user complete browser actions only for claiming a site, OAuth approval, and connecting or authorizing GitHub. Keep ownership of the terminal workflow and resume after each user action. For command flags and troubleshooting, use the CLI command reference.
  2. Navigate to your docs directory.
  3. (Optional) Add the CLAUDE.md file below to your project.
  4. Run claude to start.
Once Claude Code is running, paste the following prompt to load the Mintlify skill, search MCP, and admin MCP into the session. This gives Claude full context on Mintlify components and your site’s content.

Install the Mintlify skill and MCP servers for AI coding tools.

Open in Cursor
To install the Mintlify skills and configure the admin MCP from your terminal instead, run mint agent setup in your docs directory:
Then start Claude Code, open /mcp, and authenticate mintlify-admin. To connect only the admin MCP: Add the admin MCP server with the Claude Code CLI:
On first use, Claude Code opens a browser window to complete the OAuth login. After authenticating, Claude Code reuses the session for subsequent calls. To confirm your setup, send a prompt to Claude Code like List the Mintlify components you can use.

Template for CLAUDE.md

Save a CLAUDE.md file at the root of your docs directory to help Claude Code understand your project. This file trains Claude Code on your documentation standards, preferences, and workflows. See Manage Claude’s memory in the Anthropic docs for more information. Copy this example template or make changes for your docs specifications:

Sample prompts

Once you have Claude Code set up, try these prompts to see how it can help with common documentation tasks. You can copy and paste these examples directly, or adapt them for your specific needs.

Convert notes to polished docs

Turn rough drafts into proper Markdown pages with components and frontmatter. Example prompt:

Review docs for consistency

Get suggestions to improve style, formatting, and component usage. Example prompt:

Update docs when features change

Keep documentation current when your product evolves. Example prompt:

Generate comprehensive code examples

Create multi-language examples with error handling. Example prompt:

Extend Claude Code

Beyond manually prompting Claude Code, you can integrate it with your existing workflows.

Automate with GitHub Actions

Run Claude Code automatically when code changes to keep docs up to date. You can trigger documentation reviews on pull requests or update examples when APIs change.

Multi-instance workflows

Use separate Claude Code sessions for different tasks - one for writing new content and another for reviewing and quality assurance. This helps maintain consistency and catch issues that a single session might miss.

Team collaboration

Share your refined CLAUDE.md file with your team to ensure consistent documentation standards across all contributors. Teams often develop project-specific prompts and workflows that become part of their documentation process.

Custom commands

Create reusable slash commands in .claude/commands/ for frequently used documentation tasks specific to your project or team.

Frequently asked questions

No, but it significantly improves output quality. Without a CLAUDE.md, Claude Code works from general context and may not follow your specific style guide, component preferences, or terminology. A CLAUDE.md file trains Claude Code on your project’s standards so you don’t need to re-explain them in every prompt.
Claude Code is a command-line tool designed for agentic, multi-step tasks across an entire repository. It’s well-suited for tasks like auditing all pages for missing alt text, updating a parameter name across every code example, or reviewing a set of files for consistency. Cursor and Devin Desktop are IDE-based tools better suited for editing individual files with inline suggestions. Both approaches work. The right choice depends on whether your work is file-by-file or across the whole repo.
Yes. Claude Code can read your source code and generate corresponding documentation. Point it at an API endpoint, a configuration file, or a set of functions and ask it to produce matching documentation following your CLAUDE.md standards. Review and refine the output. Automated generation works best when you treat Claude Code as a first draft author, not a final publisher.
Commit the CLAUDE.md file to your documentation repository. Anyone who clones the repo and runs Claude Code automatically uses your project’s configuration. This makes documentation standards consistent across contributors without requiring each person to set up their own context.