Local Context7

local-context7

English | 中文

Build local documentation reference material for AI coding agents.

This repo is designed to be used with Claude Code. Open this repo in Claude Code and use slash commands (/build-my-context7, /generate-agent-skills, /install-agent-skills) to download docs, generate skills, and install them. The skills and scripts are meant to be invoked by Claude Code, not run manually.

What It Does

local-context7 downloads official documentation from various sources, filters it with AI assistance to keep only developer-relevant content, and generates skills that make the docs available as context during development.

Supported AI Agents:

• Claude Code (~/.claude/skills/)
• OpenAI Codex CLI (~/.codex/skills/)
• OpenCode (~/.config/opencode/skills/)

Included Documentation:

Quick Start

Prerequisites

• Claude Code installed
• jq, git, curl available in your shell

Usage

1. Open this repo in Claude Code
2. Run the slash commands:

/build-my-context7           # Download and filter all manifests
/generate-agent-skills       # Generate skills for all agents
/install-agent-skills        # Install skills to system locations

To process a single library:

/build-my-context7 zod-docs        # Download only zod-docs
/generate-agent-skills zod-docs    # Generate only zod-docs skill
/install-agent-skills zod-docs     # Install only zod-docs skill

After installation, the skills are available in your other projects. For example, when working on a Next.js app, Claude Code will automatically have access to nextjs-canary-docs.

Adding New Documentation

Just give Claude Code the GitHub URL of the docs folder:

"Add docs from https://github.com/prisma/docs/tree/main/content"

Claude Code will automatically create the manifest and run the build commands. This is the fastest way to add new documentation.

Or create a manifest manually:

1. Create a JSON file in .claude/skills/download-docs/scripts/manifests/
2. Run /build-my-context7 <name> → /generate-agent-skills <name> → /install-agent-skills <name>

Manifest Formats

Manifests use JSON Schema for validation. Add "$schema": "./manifest.schema.json" for IDE autocomplete and validation.

GitHub Source (for repos with docs):

{
  "$schema": "./manifest.schema.json",
  "_source": {
    "type": "github",
    "repo": "owner/repo-name",
    "branch": "main",
    "path": "docs",
    "extensions": [".md", ".mdx"],
    "exclude": ["**/internal/**"]
  }
}

URL Source (for individual files):

{
  "$schema": "./manifest.schema.json",
  "_source": { "type": "url" },
  "files": {
    "getting-started": {
      "installation": "https://example.com/docs/install.md"
    }
  }
}

URL Source with HTML Conversion (requires pandoc):

{
  "$schema": "./manifest.schema.json",
  "_source": { "type": "url", "convert": "html" },
  "files": {
    "docs": { "overview": "https://example.com/docs/overview.html" }
  }
}

Limitation: HTML conversion only works for static HTML pages. Single Page Applications (SPAs) that render content via JavaScript (e.g., React/Angular apps) will produce empty output. See TODO below for planned Playwright support.

Directory Structure

.claude/                    # Working config for this repo
├── skills/                 # Skills for building docs
│   ├── build-my-context7/  # Main orchestration skill
│   ├── download-docs/      # Downloads from manifests
│   ├── filter-docs/        # AI-assisted filtering
│   ├── generate-agent-skills/
│   └── install-agent-skills/
└── agents/                 # Sub-agents

dotfiles/                   # Generated skills (source of truth)
├── shared/                 # Documentation stored once
├── claude/skills/          # SKILL.md → ~/.claude/skills/
├── codex/skills/           # SKILL.md → ~/.codex/skills/
└── opencode/skills/        # SKILL.md → ~/.config/opencode/skills/

output/                     # Intermediate files (gitignored)

How It Works

1. Download - Fetches documentation from GitHub repos or URLs based on manifest configurations
2. Filter - AI reviews each file and removes non-useful content (changelogs, contribution guides, marketing pages, etc.)
3. Generate - Creates skills for each AI agent following their specific format requirements
4. Install - Symlinks generated skills to system locations where agents can find them

Why Local?

• Offline access - Documentation available without internet
• Version control - Pin to specific documentation versions
• Customization - Add your own documentation sources
• Privacy - No data sent to external services during development
• Speed - Instant context loading without API calls

TODO

• Playwright handler for SPA sites - Add a "type": "playwright" source handler to support JavaScript-rendered documentation sites (e.g., antigravity.google). Would use headless Chromium to render pages before extracting content.

License

This project is open source and available under the MIT License.