mjm.local.docs

A self-hosted MCP server that lets AI coding agents (Claude Code, OpenCode, etc.) search your local documents using semantic search. Add your specs, docs, or notes — your AI tools find them automatically.

Built with .NET 10, SQLite, and Blazor Server.

Prerequisites: .NET 10 SDK (10.0.102+)

Quick Start

git clone https://github.com/markjackmilian/mjm.local.docs
cd mjm.local.docs/mjm.local.docs
dotnet run --project src/Mjm.LocalDocs.Server/Mjm.LocalDocs.Server.csproj

Open http://localhost:5024 for the web UI. The MCP endpoint is at http://localhost:5024/mcp.

No API key needed by default — uses fake embeddings in development.

Connect to Your AI Agent

Add to .claude/settings.json (project) or ~/.claude/settings.json (global):

{
  "mcpServers": {
    "local-docs": {
      "type": "http",
      "url": "http://localhost:5024/mcp"
    }
  }
}

Same config works for OpenCode (opencode.json). Restart your agent after adding.

Available MCP Tools

Tool	Description
search_docs	Semantic search across all documents
add_document	Add a .txt, .md, .pdf, or .docx file
update_document	Replace a document with a new version
list_collections	List all projects
delete_document	Remove a document and its embeddings

Configuration

All settings live under the "LocalDocs" key in appsettings.json.

Embedding providers: Fake (default, no key), OpenAI, AzureOpenAI, Ollama

Storage providers: InMemory, Sqlite (default), SqliteHnsw (recommended for >10k docs), SqlServer

File storage: Database (default), FileSystem, AzureBlob

Production example — OpenAI embeddings + HNSW index:

{
  "ConnectionStrings": { "LocalDocs": "Data Source=/data/localdocs.db" },
  "LocalDocs": {
    "Embeddings": { "Provider": "OpenAI", "Dimension": 1536, "OpenAI": { "Model": "text-embedding-3-small" } },
    "Storage": { "Provider": "SqliteHnsw", "Hnsw": { "IndexPath": "/data/hnsw_index.bin" } }
  }
}

Set OPENAI_API_KEY as an environment variable — never commit API keys.

License

MIT