QdrantSkillsMCP.Models.MiniLM 1.5.4

QdrantSkillsMCP

A .NET 10 MCP server for vector-based skill storage and retrieval using Qdrant. Enables AI agents (Claude Code, Copilot, Codex, etc.) to semantically search, load, and manage skills via MCP tools.

Prerequisites

• .NET 10 SDK (required)
• A Qdrant instance — local via Docker/Aspire, or a hosted service like Qdrant Cloud

Get Started

No install needed — dnx runs the tool directly from NuGet, always using the latest version:

# Initialize config (creates ~/.qdrant-skills/config.json with local defaults)
dnx QdrantSkillsMCP -- --config init

# Auto-configure your AI agent (Claude, Copilot, Codex, etc.)
dnx QdrantSkillsMCP -- --setup

# Verify your Qdrant connection
dnx QdrantSkillsMCP -- --config validate

What is dnx? It's .NET 10's equivalent of npx — runs NuGet tools without installing them. Always gets the latest version automatically.

Alternative: Global Install

If you prefer a permanent installation (no dnx prefix needed):

dotnet tool install -g QdrantSkillsMCP
qdrant-skills-mcp --config init
qdrant-skills-mcp --setup

Update later with: dotnet tool update -g QdrantSkillsMCP

Configuration

# Show all config with source annotations ([default], [user], [project], [env])
dnx QdrantSkillsMCP -- --config show

# Connect to a remote Qdrant instance
dnx QdrantSkillsMCP -- --config set QdrantHost=my-qdrant.example.com
dnx QdrantSkillsMCP -- --config set QdrantGrpcPort=6334
dnx QdrantSkillsMCP -- --config set UseTls=true
dnx QdrantSkillsMCP -- --config set QdrantApiKey=your-api-key

# Named profiles for switching between environments
dnx QdrantSkillsMCP -- --config use cloud

# Validate connection works
dnx QdrantSkillsMCP -- --config validate

# Generate env var template for your shell (auto-detects bash/PowerShell/cmd)
dnx QdrantSkillsMCP -- --config env

# Interactive config wizard
dnx QdrantSkillsMCP -- --config

Config files:

• User-level: ~/.qdrant-skills/config.json (API keys, personal settings)
• Project-level: ./qdrant-skills.json (shared team settings)
• Precedence: Environment variables > Project > User > Defaults

CLI Usage

# Search skills by meaning
dnx QdrantSkillsMCP -- --console search "authentication patterns"

# List all skills
dnx QdrantSkillsMCP -- --console list

# JSON output for scripting
dnx QdrantSkillsMCP -- --console --json search "error handling"

# Interactive REPL with tab completion and history
dnx QdrantSkillsMCP -- --console

# Show help
dnx QdrantSkillsMCP -- --console help

MCP Server Mode

By default (no flags), QdrantSkillsMCP runs as an MCP server over stdio. This is how AI agents connect to it. The --setup wizard configures this automatically for your agent.

Available MCP Tools

ONNX Model Packages

For local embedding without API keys, three pre-built model packages are available:

Without a companion package, the tool auto-downloads all-MiniLM-L6-v2 from HuggingFace on first use (~23 MB, requires internet). This works with no setup.

To pre-install a model, add it to your NuGet global cache (works without a project):

# Create a temp project, restore, then discard — populates the NuGet cache
dotnet new console -o /tmp/model-install --no-restore
dotnet add /tmp/model-install package QdrantSkillsMCP.Models.BgeSmall
dotnet restore /tmp/model-install

The tool auto-detects the model in ~/.nuget/packages/qdrantskillsmcp.models.bgesmall/ on startup.

Then select it:

dnx QdrantSkillsMCP -- --config set OnnxModelName=bge-small-en-v1.5

Embedding Providers

Configure via dnx QdrantSkillsMCP -- --config set EmbeddingProvider=<provider>:

Development

Requires .NET 10 SDK and Docker (for Qdrant via Aspire).

# Run with Aspire (starts Qdrant automatically)
dotnet run --project src/QdrantSkillsMCP.AppHost

# Run unit tests
dotnet test tests/QdrantSkillsMCP.UnitTests

# Run all tests (requires Qdrant running)
dotnet test

License

MIT