context-bridge: Instructions file for Claude Code

ContextBridge — AI Assistant Context

Project Overview

ContextBridge is a Windows-first, zero-dependency MCP memory server. It gives AI coding assistants (Claude Code, Claude Desktop, and others) a shared persistent memory layer backed by semantic vector search. A single Windows Service process holds the embedding model warm and serves all concurrent clients via Streamable HTTP MCP transport.

Full design rationale and architectural decisions: docs/DESIGN.md

NuGet package ID: ThunderEagle.ContextBridge
Code namespaces: ContextBridge.*

---

Solution Structure

src/
  ContextBridge.Core/           # Domain interfaces, models — no infrastructure deps
  ContextBridge.Infrastructure/ # SQLite, ONNX Runtime, MCP tool implementations
  ContextBridge.Service/        # Single executable: Worker Service host + CLI entry point
  ContextBridge.Cli/            # Class library: CLI command handler implementations
tests/
  ContextBridge.Tests/          # xUnit integration tests
docs/
  DESIGN.md                     # Full design document
  adr/                          # Architectural Decision Records
models/                         # Bundled ONNX model files (binary, committed intentionally)

Single executable model: ContextBridge.Service is the only executable. It bootstraps System.CommandLine at startup — if CLI args are present it routes to a command handler in ContextBridge.Cli and exits; otherwise it starts the Worker Service / Windows Service host. See ADR-002.

Dependency direction:

ContextBridge.Service (exe)
├── ContextBridge.Cli (lib)
├── ContextBridge.Infrastructure (lib)
└── ContextBridge.Core (lib)

ContextBridge.Cli (lib)
├── ContextBridge.Infrastructure (lib)
└── ContextBridge.Core (lib)

Core has no external package dependencies.

---

Technology Stack

Concern	Technology
Runtime	.NET 10 Web/Worker Service (Microsoft.NET.Sdk.Web)
Windows Service	Microsoft.Extensions.Hosting.WindowsServices
Embeddings	Microsoft.ML.OnnxRuntime + all-MiniLM-L6-v2 INT8 ONNX
AI abstraction	Microsoft.Extensions.AI (IEmbeddingGenerator<string, Embedding<float>>)
Storage	Microsoft.Data.Sqlite + sqlite-vec extension
Data access	Dapper (raw SQL, Dapper mapping — EF Core avoided; sqlite-vec requires raw SQL)
MCP transport	Streamable HTTP via MCP C# SDK (ModelContextProtocol)
Secret storage	Microsoft.AspNetCore.DataProtection (cross-platform; DPAPI on Windows)
CLI parsing	System.CommandLine