Rag.NET.Mcp.Tool 1.0.0

Rag.NET.Mcp.Tool

A self-contained Model Context Protocol server for Rag.NET, packaged as a .NET global tool (ragnet-mcp) — run a RAG-backed MCP server from configuration alone, no C# project required.

Install

dotnet tool install -g Rag.NET.Mcp.Tool

Configure

The tool wires its pipeline — chat client, embedding generator, vector store — from an appsettings.json next to its working directory, or from environment variables (using __ as the section separator, e.g. RagNet__ChatClient__ApiKey) — the standard .NET configuration layering WebApplication.CreateBuilder(args) already provides. A sample appsettings.sample.json, showing every supported VectorStore:Kind, ships alongside the installed binaries; copy it to appsettings.json and fill in real values.

{
  "RagNet": {
    "ChatClient": {
      "Endpoint": "https://api.openai.com/v1",
      "ApiKey": "…",
      "Model": "gpt-4o-mini"
    },
    "Embeddings": {
      "Endpoint": "https://api.openai.com/v1",
      "ApiKey": "…",
      "Model": "text-embedding-3-small",
      "VectorDimensions": 1536
    },
    "VectorStore": {
      "Kind": "InMemory | Qdrant | PgVector",
      "Qdrant": { "Host": "…", "Port": 6334, "CollectionName": "…" },
      "PgVector": { "ConnectionString": "…" }
    }
  }
}

The chat client and embedding generator both go through one OpenAI-compatible endpoint — that covers OpenAI, Azure OpenAI, OpenRouter, Ollama, and LM Studio, since they all speak the same wire API. The vector store is one of three kinds: InMemory (the default, zero setup, but every ingested document is lost when the process exits — a warning is logged at startup), Qdrant, or PgVector.

A misconfigured setting — an unrecognised Kind, a missing Endpoint/Model, an absent or non-positive VectorDimensions — fails at startup with a message naming both the setting and the configuration key that fixes it, rather than failing the first time an MCP client calls a tool.

Need a provider outside that set — Weaviate, Pinecone, Chroma, Azure AI Search, ONNX embeddings, or a bespoke IChatClient? Host the Rag.NET.Mcp library directly in your own application and register whatever you like; this tool covers the bounded, OpenAI-compatible set above and nothing wider.

Run

# stdio transport (Claude Desktop subprocess)
ragnet-mcp

# HTTP/SSE on port 5050 with API-key auth
ragnet-mcp --transport http --port 5050 --api-key your-secret

# Behind a gateway that already authenticates — has to be said out loud
ragnet-mcp --transport http --port 5050 --allow-anonymous

The HTTP transport refuses to start without an authentication decision. It exposes ingest, retrieve and ask, and binds every interface rather than loopback — so --transport http with neither --api-key (or RAGNET_MCP_API_KEY) nor --allow-anonymous exits with a message naming both ways out, rather than serving your store to the network. --allow-anonymous starts and logs a warning each run.

--allow-anonymous never weakens a key you did configure: supply both and the key still decides, so a stray flag in a script cannot quietly disable authentication.

Claude Desktop configuration for the stdio variant:

{
  "mcpServers": {
    "ragnet": { "command": "ragnet-mcp" }
  }
}

The server exposes the rag_retrieve, rag_ask and rag_ingest tools to the MCP host. Hosting the server inside your own application instead? Use the Rag.NET.Mcp library package.

Full guide

• MCP server