Legible Connect: Setup Guide for Claude Code & OpenAI Codex Integration

Prerequisites

• A Legible account with at least one connected site.
• Your workspace ID — found in your dashboard under Settings.
• An API key — generate one from Settings → API Keys.

Connect to Claude Code

Open your Claude Code MCP configuration. The config location depends on your setup: global at ~/.claude/claude_desktop_config.json, or per-project at .claude/mcp.json in your project root.

Add the Legible MCP server configuration:

{
  "mcpServers": {
    "legible": {
      "url": "https://mcp.getlegible.com/YOUR_WORKSPACE_ID",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Connect to OpenAI Codex

Codex uses the same MCP configuration format. Add the Legible server to your Codex MCP settings:

{
  "mcpServers": {
    "legible": {
      "url": "https://mcp.getlegible.com/YOUR_WORKSPACE_ID",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Available Tools

Legible Connect exposes three MCP tools that your AI agent can call automatically:

• search_context — Search your knowledge base for relevant content. Parameters: query (string, required), limit (number, optional, default 5). Returns content chunks with relevance scores and source metadata.
• get_document — Retrieve a full document by its ID. Parameters: document_id (string, required). Returns the full document in Markdown with metadata.
• list_sources — List all content sources in your workspace. No parameters. Returns an array of sources with document counts and last-updated timestamps.

How It Works in Practice

When you ask your AI agent something like 'Write a blog post about our new pricing model,' the agent automatically calls search_context to find relevant content from your knowledge base — existing blog posts, pricing pages, brand guidelines. The response draws from YOUR content, not generic training data.

This means on-brand copy, accurate product details, and consistent messaging — all without manually pasting context into every prompt.

Endpoint Reference

• Base URL: https://mcp.getlegible.com/{workspace_id}
• Authentication: Bearer token via Authorization header.
• Protocol: MCP (Model Context Protocol) over HTTPS.
• Rate limits: 100 requests/minute (configurable per plan).

Troubleshooting

• "Connection refused" — Check your workspace ID and API key are correct.
• "No results returned" — Ensure your site has been crawled and content is indexed. Check the Content Library in your dashboard.
• "Rate limit exceeded" — Upgrade your plan or reduce request frequency.
• "Unauthorized" — Regenerate your API key from Settings → API Keys.
• Agent doesn't use Legible tools — Restart your AI agent after adding the MCP config.