Using jiki with AI Tools

AI coding assistants work best when they have accurate, up-to-date context about the libraries you’re using. jiki provides three structured context files specifically designed for AI consumption.

Context files

When to use which:

• Quick questions (“how do I write a file?”) — llm.txt has enough context
• Implementation help (“set up a Vite dev server with HMR and plugins”) — use llm-full.txt
• Automated discovery — llms.txt is the entry point for tools that follow the llms.txt spec

Using with Claude Code

Add jiki context to your project’s CLAUDE.md file:

## jiki

This project uses jiki for browser-based code execution.
Full API reference: https://jiki.sh/llm-full.txt
Limitations (important): https://jiki.sh/docs/getting-started/limitations

Or fetch the context directly in a conversation:

@https://jiki.sh/llm-full.txt

Using with Cursor

Add jiki documentation as project context. Create or update .cursorrules:

## jiki Context

This project uses jiki (browser-based Node.js runtime).
Refer to https://jiki.sh/llm-full.txt for the complete API reference.

Key constraints:
- No native addons or .node files
- No TCP/UDP sockets — virtual HTTP servers only
- In-memory filesystem — use persistence adapter for durability
- esbuild-wasm for transpilation (3-10x slower than native)

You can also add llm-full.txt as a docs reference in Cursor’s project settings.

Using with GitHub Copilot

Reference jiki context in Copilot Chat:

@workspace Use the jiki API from https://jiki.sh/llm.txt to help me set up a React playground

For Copilot’s workspace agent, having llm.txt or llm-full.txt in your project’s documentation folder helps it discover jiki’s API surface.

Using as MCP context

If your AI tool supports the Model Context Protocol, you can serve jiki docs as an MCP resource:

// Example: fetch jiki context for an MCP server
const response = await fetch("https://jiki.sh/llm-full.txt");
const context = await response.text();

// Provide as a resource to the AI
server.addResource({
  uri: "jiki://docs/full",
  name: "jiki Full Documentation",
  mimeType: "text/plain",
  text: context,
});

Tips for better AI assistance

1. Start with llm.txt for quick lookups, switch to llm-full.txt when you need the AI to write implementation code.

2. Point at specific sections if you’re working on one feature. The llm-full.txt file is organized by topic — tell the AI “focus on the Plugins API section” to reduce noise.

3. Include limitations early in your context. The single most common AI mistake with jiki is suggesting Node.js features that don’t exist in the browser runtime (native addons, raw sockets, fs.promises completeness).

4. Use the examples as starting points. Tell your AI “adapt the pattern from the Claude coding example” — the example code is referenced in llm-full.txt.

5. Version awareness. The context files are updated with each release. If you’re on an older version, the current llm-full.txt may reference APIs that don’t exist in your version yet.

Next steps

• Provider Setup — configure API access for building AI-powered apps
• Building an AI Coding Assistant — patterns for streaming, code injection, and error recovery