FieldCure.Mcp.Rag 2.4.1

FieldCure.Mcp.Rag

Requires Ollama 0.4.0 or later when using Ollama for embedding or contextualization.

MCP RAG server with hybrid BM25 + vector search and AI-powered chunk contextualization — indexes documents from configured source paths, enriches chunks with AI-generated context and keywords, generates embeddings, and performs keyword (FTS5) and semantic (cosine similarity) search with Reciprocal Rank Fusion.

Install

dotnet tool install -g FieldCure.Mcp.Rag

Commands

fieldcure-mcp-rag
├── serve         --base-path <path>                         # Multi-KB MCP search server (stdio)
├── exec          --path <kb-path> [--force] [--partial ...]  # Headless indexing
├── exec-queue    --queue-file <path> [--sweep-all]           # Sequential queue orchestrator
└── prune-orphans --base-path <path>                         # Delete orphan KB folders

• serve — read-only MCP server serving all KBs under the base path. Lazy-loads per KB.
• exec — headless indexing with 2-commit model, binary-split failure isolation, deferred retry.
• exec-queue — sequential orchestrator for queued indexing requests. No GPU contention.
• prune-orphans — deletes GUID-named folders without config.json. Protects backups.

API keys:

• serve (stdio) — environment variable (OPENAI_API_KEY, ANTHROPIC_API_KEY, etc.) → MCP Elicitation fallback on the first tool call that needs a key. Session cache, max 2 re-elicits.
• exec / exec-queue (headless batch) — environment variable only. If unset, the run soft-fails with a clear message.

Quick Start

{
  "id": "my-kb-001",
  "name": "Project Docs",
  "sourcePaths": ["C:\\Users\\me\\Documents\\project-docs"],
  "embedding": {
    "provider": "openai",
    "model": "text-embedding-3-small",
    "apiKeyPreset": "OpenAI"
  }
}

# Index
fieldcure-mcp-rag exec --path "%LOCALAPPDATA%\FieldCure\Mcp.Rag\my-kb-001"

# Serve
fieldcure-mcp-rag serve --base-path "%LOCALAPPDATA%\FieldCure\Mcp.Rag"

Claude Desktop

{
  "mcpServers": {
    "rag": {
      "command": "fieldcure-mcp-rag",
      "args": ["serve", "--base-path", "C:\\Users\\me\\AppData\\Local\\FieldCure\\Mcp.Rag"],
      "env": {
        "OPENAI_API_KEY": "sk-..."
      }
    }
  }
}

Tools (7)

config.json Reference

Supported Formats

DOCX, HWPX, XLSX, PPTX, PDF, TXT, MD. Scanned PDFs without a text layer fall back to Tesseract OCR on Windows only — see "Platform support" below.

Platform support

Cross-platform on Windows, Linux, macOS. Text extraction from all supported document formats works everywhere. The optional OCR package (FieldCure.DocumentParsers.Ocr, ships Tesseract native binaries) is referenced conditionally in the server's .csproj via $([MSBuild]::IsOSPlatform('Windows')), so Linux and macOS builds are pure managed code and scanned-PDF pages on those platforms yield empty text.

Requirements

• .NET 8.0 Runtime or later

See Also

Part of the AssistStudio ecosystem.

• GitHub
• MCP Specification