llmctx

svelte-llm

LLM presets for Svelte 5 and SvelteKit. Visit the site at svelte-llm.khromov.se.

MCP Endpoint

This service provides an MCP (Model Context Protocol) endpoint for use with AI assistants:

• SSE (Claude Desktop): https://svelte-llm.khromov.se/mcp/sse
• Streamable HTTP (most other clients): https://svelte-llm.khromov.se/mcp/mcp

OG image from https://dynamic-og-image-generator.vercel.app/

Setup

• Create a .env file with the content GITHUB_TOKEN=
• Create a Classic GitHub token. It must have public_repo permissions.
• Enter this in the .env file.

llmctx

llmctx transforms technical documentation into AI-ready formats. It provides a simple way to access condensed, LLM-friendly versions of popular framework and library documentation through preset URLs. This makes it easier for developers to use AI coding assistants like Cursor or Zed with accurate, up-to-date context from official documentation sources.

• Preset URLs for quick access to LLM-optimized documentation
• Regularly updated content from official documentation

Supported presets

Visit llmctx.com to see all presets.

Adding presets

To add a new preset:

1. Fork this repo.

2. Open the src/lib/presets.ts file.

3. Add a new entry to the presets object with the following structure:

   [presetKey]: {
     title: 'Preset Title',
     owner: 'github-owner',
     repo: 'github-repo',
     glob: ['**/*.md', '**/*.mdx', '!**/excluded/**'], // Required, supports glob patterns
     prompt: 'Optional prompt for additional context'
   }
   

4. Create a pull request with your changes.

The glob field supports glob patterns, providing flexible file matching:

• Use **/*.md to match all Markdown files in any subdirectory
• Use **/*.mdx to match all MDX files in any subdirectory
• Use !pattern to exclude files/directories matching the pattern
• Combine patterns for fine-grained control, e.g., ['**/*.md', '**/*.mdx', '!**/excluded/**']
• Patterns are processed in order, so you can include files and then exclude specific ones

Please ensure that the documentation source is reliable and actively maintained.

Debug MCP

NODE_TLS_REJECT_UNAUTHORIZED=0 NODE_OPTIONS="--insecure-http-parser" npx @modelcontextprotocol/inspector

http://127.0.0.1:6274

You can also use GitHub Copilot in Agent mode to try the agent locally, see .vscode/mcp.json for info on how that works.