Skip to content

optional-features.md

Latest commit

 

History

History
64 lines (43 loc) · 1.99 KB

optional-features.md

File metadata and controls

64 lines (43 loc) · 1.99 KB

Optional Features

CodeForge includes several features that are available but not enabled by default. This guide covers how to enable and configure them.

mcp-qdrant (Vector Memory for Claude)

Adds persistent vector memory to Claude Code via a Qdrant MCP server. Claude can store and retrieve information across sessions.

Enabling

Add to devcontainer.json under "features":

"./features/mcp-qdrant": {
    "collectionName": "my-project-memory",
    "embeddingModel": "all-MiniLM-L6-v2"
}

Options

Option Default Description
collectionName agent-memory Qdrant collection name
embeddingModel all-MiniLM-L6-v2 Embedding model for vector search
qdrantUrl (empty) Remote Qdrant server URL. If empty, uses local storage.
qdrantApiKey (empty) API key for remote Qdrant server
qdrantLocalPath /workspaces/.qdrant/storage Local storage path (when no URL set)

Supported Embedding Models

  • all-MiniLM-L6-v2 (default, smallest, fastest)
  • BAAI/bge-small-en-v1.5
  • BAAI/bge-base-en-v1.5
  • sentence-transformers/all-mpnet-base-v2

Prerequisites

Already met by default container: Python 3.14 and uv are pre-installed.

How It Works

  1. During container build, the embedding model is pre-downloaded from GCS (not HuggingFace, to avoid network issues in containers).
  2. On container start, a post-start hook registers the Qdrant MCP server in Claude Code's settings.json.
  3. Claude Code can then use qdrant-store and qdrant-find tools to persist and search memories.

Verification

uvx mcp-server-qdrant --help

Disabling Default Features

Any feature can be disabled without removing it from devcontainer.json by setting "version": "none":

"./features/hadolint": { "version": "none" },
"./features/shellcheck": { "version": "none" }

The feature entry stays in the config for easy re-enabling — just remove "version": "none" or set it to "latest".