dude-claude-plugin

Ultra-minimal RAG and cross-project memory for Claude Code.

Dude gives Claude a persistent memory across projects. It stores issues and specifications in a local SQLite database with vector embeddings, so Claude automatically recalls relevant context from past sessions.

Install

From marketplace (recommended)

Installs MCP tools and hooks (auto-retrieve on prompt, auto-persist on stop):

claude plugin marketplace add fingerskier/claude-plugins
claude plugin install dude-claude-plugin@fingerskier-plugins

MCP server only (via npx)

If you just want the 7 MCP tools without auto-hooks:

claude mcp add dude -- npx dude-claude-plugin mcp

Global install

npm install -g dude-claude-plugin
claude mcp add dude -- dude-claude mcp

What it does

| Component | Description | |-----------|-------------| | MCP server | 7 tools: search, upsert_record, get_record, list_records, delete_record, list_projects, sync_status | | Auto-retrieve hook | On each prompt, searches memory for relevant context and injects it | | Auto-persist hook | After each response, classifies the work and saves issues/specs | | Web UI | Local dashboard at http://127.0.0.1:3456 for manual CRUD | | Storage | libsql (SQLite-compatible) at ~/.dude-claude/dude-libsql.db | | Embeddings | Local all-MiniLM-L6-v2 via @huggingface/transformers (no API keys) | | Cloud sync | Optional Turso cloud sync via environment variables |

How it works

1. You submit a prompt -- the auto-retrieve hook embeds it and searches for related records across all your projects
2. Matching context is injected so Claude has relevant history before reasoning
3. After Claude responds, the auto-persist hook classifies the work:
   • Bug fix? Upserts an issue record (status: resolved)
   • Improvement? Upserts a spec record
   • Neither? Skips silently
4. Next session, step 1 finds those records automatically

Web UI

npx dude-claude-plugin serve
# or if globally installed:
dude-claude serve

Opens a local dashboard at http://127.0.0.1:3456 (auto-opens in your browser). The UI has two panels: a sidebar listing records with filters, and a main panel for viewing and editing.

What you can do:

• Project selector — filter records by project or view all projects at once
• Semantic search — type in the search bar to find records by meaning, not just keywords (results ranked by similarity)
• Filter by kind — narrow to issues, specs, arch decisions, updates, or tests
• Filter by status — open, resolved, archived, active, or inactive
• Create / edit / delete — click "+ New" to add a record, click any record to edit it, or delete from the detail view

Set DUDE_PORT to change the default port (see Configuration).

Configuration

| Env variable | Default | Description | |---|---|---| | DUDE_PORT | 3456 | Web UI port | | DUDE_CONTEXT_LIMIT | 5 | Max records injected per prompt | | DUDE_RECENCY_HOURS | 1 | Lookback window for recent records |

Cloud Sync (optional)

The plugin works fully offline by default. To enable cloud sync with Turso, set these environment variables:

| Env variable | Description | |---|---| | DUDE_TURSO_URL | Turso database URL (e.g. libsql://your-db.turso.io) | | DUDE_TURSO_TOKEN | Turso auth token | | DUDE_SYNC_INTERVAL | Sync interval in ms (default: 60000) |

When configured, @libsql/client maintains a local embedded replica that auto-syncs with Turso. You can also trigger a manual sync via the sync_status MCP tool or POST /api/sync.

Requirements

• Node.js >= 18

License

MIT