MCP Server

In the age of generative AI, documentation is no longer just for humans—it's for agents too. Lithos includes a native Model Context Protocol (MCP) server, making your entire knowledge base instantly accessible to AI assistants like Claude, Cursor, and Visual Studio Code.

Native Model Intelligence

Lithos treats AI integration as a first-class citizen, not an afterthought.

The MCP Standard

The Model Context Protocol is an open standard that defines how AI models interact with external data. By adopting this standard, Lithos ensures that your documentation is not just a static website, but a dynamic Context Server. This means any MCP-compliant AI client can "plug in" to your documentation and read it in real-time.

Zero-Configuration

For most users, there is absolutely nothing to configure. The MCP server starts automatically alongside your Docus dev server.

• Production Ready: When deployed, the MCP endpoints are served alongside your static assets.
• Auto-Discovery: Clients can query the server capabilities without manual mapping.
• Seamless Updates: As you update your markdown files, the AI context is updated instantly.

Privacy & Control

Since Lithos processes your Obsidian vault locally, your interaction with local AI models remains private. You decide what is exposed. The API is read-only by default, ensuring that AI agents can learn from your docs but cannot hallucinate changes or delete your managed content.

Semantic Integration

We don't just dump text into the context window; we structure it for semantic understanding.

Proactive Tool Use

When connected, the AI doesn't just "know" your docs exists; it has tools to explore it.

• list_pages: The Agent can scan your directory structure to understand the scope of available topics.
• get_page: The Agent can fetch specific content when it identifies a relevant file.
• search_docs: The Agent can perform semantic searches to find answers to user questions. This "active retrieval" pattern is far superior to simply pasting content into a chat window, as the AI decides what it needs based on the conversation flow.

llms.txt Generation

For lighter-weight integrations or web crawling, Lithos automatically generates a standard /llms.txt file at the root of your site. This file provides a curated, token-optimized summary of your documentation, designed specifically for training data or quick context ingestion by simpler models that don't support full MCP.

Context Awareness

Because Lithos understands the structure of your vault (links, hierarchy, tags), it preserves this metadata in the API context. When an AI reads a page, it also sees the relationships—what links here, what tags are used—allowing for much deeper reasoning. An AI can follow a trail of wikilinks just like a human reader would, hopping from concept to concept to build a complete answer.

Ecosystem Compatibility

Lithos is designed to play well with the tools you already use.

Major Clients Supported

We test against the most popular AI coding environments:

• Claude Desktop: Connects directly via the MCP settings for full context in chat.
• Cursor: Use your docs as a @doc source for codebase indexing and RAG.
• Visual Studio Code: Integrate with Copilot Chat or other agents to query your docs while you code.
• Windsurf & Zed: Fully supported via their respective MCP configuration panels.

Configuration Guide

Connecting to your local Lithos instance is usually as simple as adding a JSON configuration.

{
  "mcpServers": {
    "my-docs": {
      "type": "http",
      "url": "http://localhost:3000/mcp"
    }
  }
}

This simple config tells your editor: "There is an intelligent documentation server running here."

Extending the Server

For developers, the built-in server is extensible. Using the underlying runtime, you can add custom tools. Imagine a tool that not only reads your docs but executes code snippets from them, or checks the status of a project defined in your frontmatter. The possibilities for "Executable Documentation" are built right into the architecture.