CaveauAI MCP Server

What Is It

The CaveauAI MCP Server is a standards-compliant Model Context Protocol server that exposes your CaveauAI document corpus as a set of tools any MCP-compatible AI agent can call. Instead of copying content into prompts manually, agents connect directly to your corpus and retrieve grounded, source-cited answers on demand.

How It Works

The server runs as a local process or remote endpoint and registers three core MCP tools:

• corpus_search — Semantic vector search across all documents in a corpus. Returns ranked passages with source citations, page numbers, and relevance scores.
• corpus_chat — RAG-powered conversational query against a corpus. The server retrieves relevant context, augments the prompt, and returns a grounded answer with inline citations.
• corpus_documents — List, filter, and retrieve metadata for all documents in a corpus. Useful for agents that need to understand what knowledge is available before querying.

Architecture

The MCP server communicates with the CaveauAI API over HTTPS using token-scoped authentication. All document processing, embedding generation, and inference happen on Blue Note Logic’s EU-hosted bare-metal infrastructure. No document content is sent to third-party services.

The server supports both stdio transport (for local IDE integrations like Claude Desktop and Cursor) and SSE transport (for remote and multi-user deployments). Configuration is a single JSON block in your MCP client settings.

Quick Start

Add the following to your MCP client configuration (e.g. claude_desktop_config.json):

{
  "mcpServers": {
    "corpusai": {
      "command": "npx",
      "args": ["-y", "@corpusai/mcp-server"],
      "env": {
        "CORPUSAI_API_KEY": "your-api-key",
        "CORPUSAI_CORPUS_ID": "your-corpus-id",
        "CORPUSAI_BASE_URL": "https://ai.bluenotelogic.com/api/v2"
      }
    }
  }
}

Supported Clients

• Claude Desktop (Anthropic)
• Cursor IDE
• Windsurf IDE
• VS Code with GitHub Copilot
• Continue.dev
• Any MCP-compatible agent framework

Use Cases

• Legal research — Agents query Norwegian legal corpuses built from Lovdata, returning cited statute references and case law.
• Enterprise knowledge base — Connect Copilot Studio or internal agents to company policies, SOPs, and technical documentation.
• Due diligence — Financial analysts use AI agents to search across uploaded contracts, annual reports, and regulatory filings.
• Developer documentation — Engineering teams connect their IDE agent to internal API docs, runbooks, and architecture decision records.