AI Documentation Generation: Auto-Generate Docs from Code

Four-Step Workflow: Code Analysis to Publication

The following workflow uses GitHub MCP and JSDoc/TSDoc MCP Skill as the primary tools. The same four steps apply when substituting Notion MCP for publication or README Generator MCP for new project setup.

Step 1: Code Analysis

Configure GitHub MCP and JSDoc/TSDoc MCP Skill in your assistant settings. Restart your assistant and instruct it to analyse the repository:

// ~/.claude/settings.json
{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "$GITHUB_PAT"
      }
    },
    "jsdoc-skill": {
      "command": "npx",
      "args": ["-y", "mcp-jsdoc-skill"]
    },
    "notion": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-notion"],
      "env": {
        "NOTION_API_TOKEN": "$NOTION_API_TOKEN"
      }
    }
  }
}

Step 2: Extract Types and Signatures

Ask the agent to identify undocumented exports: "Read the TypeScript source files in the src/ directory and list all exported functions, classes, and types that have no JSDoc comment or have an outdated comment that does not match the current signature." The agent produces a prioritised list of documentation gaps, sorted by how frequently each export is imported across the codebase.

Step 3: Generate Documentation

Instruct the agent to write the documentation: "For each item on the list, generate a JSDoc comment block that includes @param, @returns, @throws, and at least one @example derived from the test file for that module. Write the comments directly into the source files." The agent uses the JSDoc/TSDoc MCP Skill to update each file in place, preserving existing formatting and indentation.

Step 4: Publish

After generating inline comments, publish a human-readable API reference: "Read all updated source files, extract the JSDoc comments, and create a new Notion page under the Engineering wiki titled API Reference with a section for each module." The agent uses Notion MCP to create or update the wiki page. For static site generators, use Markdown Skill MCP to write Markdown files to the docs directory instead.

Use Cases

AI documentation generation with MCP skills covers scenarios from new project bootstrapping to ongoing maintenance documentation for mature codebases. Here are four concrete examples.

New Project README

When starting a new open-source project: "Analyse this repository directory and generate a README.md with a project description, installation instructions, usage examples pulled from the test files, and a contributing guide." The README Generator MCP reads package.json for project metadata and tests for usage patterns, producing a README that accurately reflects what the project actually does rather than a generic template.

API Reference for a Public SDK

For a TypeScript SDK with hundreds of exported types: "Read all .ts files in the src/public-api directory, generate JSDoc comments for every export, and then produce a Markdown API reference grouped by module." The agent handles generics, union types, and overloaded function signatures that are tedious to document by hand, producing reference documentation that would take days to write manually in hours.

Changelog Generation from PR History

Ask GitHub MCP to compile a changelog: "Read all merged pull requests between the v2.0.0 and v2.1.0 tags, group them by type (feature, fix, breaking change) based on their labels and titles, and write a CHANGELOG.md entry in Keep a Changelog format." This turns PR hygiene into automated release documentation with no extra effort from the release manager.

Continuous Documentation in PRs

Set up a documentation review step: "When I share a list of changed files from a pull request, identify any exported symbols whose signatures changed, regenerate their JSDoc comments, and post the updated comments as a review suggestion." This pattern keeps documentation in sync with every code change rather than letting drift accumulate over time.

Comparison Table

Use this table to match each documentation skill to your workflow. The key decision criteria are whether you need inline comments or published docs, and whether your publication target is a Markdown file, a static site, or a Notion wiki.

Frequently Asked Questions