Potluck
БесплатноНе проверенPrivacy-first personal knowledge database for your AI that ingests data exports (Google Takeout first) and exposes them via MCP tools like search and list items
Описание
Privacy-first personal knowledge database for your AI that ingests data exports (Google Takeout first) and exposes them via MCP tools like search and list items. Runs locally with no cloud or telemetry.
README
Privacy-first personal knowledge database for your AI — local-first, MCP-native.
Potluck ingests your data exports (Google Takeout first: Keep + Gmail; more sources each phase), stores everything in one local SQLite database, and exposes it three ways — a web app, a CLI, and MCP for AI assistants. No cloud, no telemetry: nothing leaves your machine.
Status — v1 rewrite in progress
| Phase | Shippable increment | Status |
|---|---|---|
| P0 Reset & Walking Skeleton | potluck serve zero-config: SPA shell; CLI/API/MCP answer stats from one service layer; bench rig + CI |
✅ v1.0.0-alpha.1 |
| P1 Storage Core & First Ingest | Google Keep from a Takeout archive (zip/tgz/dir); FTS search via CLI + MCP | ✅ v1.0.0-alpha.2 |
| P2 Gmail at Scale & Search v1 | Multi-GB mbox ingested incrementally; filtered/snippeted search | ✅ v1.0.0-alpha.3 |
| P3 MVP Interfaces | Real search/item/imports UI, MCP toolset v1 — beta.1 = MVP | ✅ v1.0.0-beta.1 |
| P4 Source Expansion & Automation | Remaining planned sources, watch-folder, scheduled GDrive pull | — |
| P5 Semantic Search | Unified embedding space, HNSW index, hybrid RRF | — |
| P6 Vision & Media Enrichment | OCR, image embeddings, faces, media gallery | — |
| P7 Linkers, People & Timeline | Related-items everywhere, people review, timeline | — |
| P8 Hardening & 1.0 | Docs, backup/restore, doctor, perf sweep | — |
Full plan, architecture, and locked decisions: pinned issue #98. v0 is archived at archive/v0.
Quickstart — 60 seconds to running
Requires uv (Python is fetched automatically).
1 — Start the server. Straight from GitHub:
uvx --from git+https://github.com/DoubleGremlin181/potluck potluck serve
CLI, API, and MCP are fully functional in this form; only the web app is not
bundled (the page at / says so and points at the API docs). Release wheels
ship with the web app embedded:
uvx --from https://github.com/DoubleGremlin181/potluck/releases/download/v1.0.0-beta.1/potluck-1.0.0b1-py3-none-any.whl potluck serve
Or Docker (web app included; data persists in the potluck-data volume):
docker run -p 127.0.0.1:8765:8765 -v potluck-data:/data ghcr.io/doublegremlin181/potluck:1.0.0-beta.1
2 — Import your data. Feed it a Google Takeout
export (Keep + Gmail today; zip, tgz, or unpacked dir) — reuse the same
uvx --from … potluck prefix for every command, or upload from the web app's
Imports page:
uvx --from git+https://github.com/DoubleGremlin181/potluck potluck import ~/Downloads/takeout-20260707T120000Z-001.zip
3 — Search it.
uvx --from git+https://github.com/DoubleGremlin181/potluck potluck search "garden fence"
The web app lives at http://127.0.0.1:8765 — potluck serve opens it for
you. Search and browse everything, start imports, and watch their progress
live.
[!WARNING] Potluck is localhost-only by design — there is no authentication in v1. Do not expose it to other machines or the internet.
MCP for AI assistants
stdio (Claude Desktop, Claude Code, …):
{
"mcpServers": {
"potluck": {
"command": "uvx",
"args": ["--from", "git+https://github.com/DoubleGremlin181/potluck", "potluck", "mcp"]
}
}
}
Streamable HTTP instead: potluck serve exposes the same tools at
http://127.0.0.1:8765/mcp — one server surface for web app, API and MCP.
Toolset: search (keyword search with ranked, snippeted hits), list_items
(browse/filter without a query), get_item (full content by id), get_thread
(the whole email conversation around an item), get_stats (database
overview), list_sources (what this build can ingest).
Setup guides — Claude Desktop, Claude Code, OpenClaw, Hermes, direct REST without MCP, and troubleshooting: docs/ai-integration.md.
CLI
potluck import PATH ingest an export (Google Takeout zip/tgz/dir; auto-detected)
potluck search Q full-text search (--kind, --prefix, --cursor, --limit, --json)
potluck list browse items without a query (--kind, --source, --since, --sort, --json)
potluck show ID full item content + metadata (--thread: the whole conversation)
potluck status database overview + per-import stats
potluck serve web app + API + MCP (/mcp) on one port (opens your browser)
potluck mcp MCP server on stdio (HTTP lives at /mcp on the serve port)
potluck bench run benchmark harness (smoke/full tiers)
potluck dev source-plugin scaffolding (new-source / check-source)
Search queries can carry inline operators, combinable with each other and
free text: from:[email protected] (or a name prefix: from:alice),
source:gmail, kind:email, after:2024-01-01 (inclusive),
before:2025-06-30 (exclusive), with quoted values supported. Source values
are case-insensitive and spaces map to underscores, so source:"google keep"
≡ source:google_keep. Invalid operator values are ignored — the response's
warnings list says what was dropped — and unknown key:value pairs are
searched as plain text. Operators alone (no free text) list the matching
items newest-first.
Search-as-you-type: --prefix (CLI) / prefix=true (MCP) matches the last
word as a prefix (gar finds garden/garlic/garnet). Pagination uses opaque
keyset cursors — pass a response's next_cursor back as cursor for the
next page; the result set is frozen at the first page, so items ingested
mid-pagination never shift, repeat, or hide existing hits.
Configuration is optional: defaults work out of the box. Override via POTLUCK_* env vars
or ~/.config/potluck/config.toml (env > toml > defaults); the database lives at
~/.local/share/potluck/potluck.db by default.
Development
git clone https://github.com/DoubleGremlin181/potluck && cd potluck
uv sync && uv run pre-commit install
(cd web && npm ci && npm run build)
uv run potluck serve
Tests: uv run pytest (unit tier), -m browser for the Playwright smoke. Conventions live
in CLAUDE.md; test patterns in tests/README.md.
License
Установить Potluck в Claude Desktop, Claude Code, Cursor
unyly install potluckСтавит в Claude Desktop, Claude Code, Cursor и VS Code — сам разбирается с npx, uvx и сборкой из исходников.
Впервые? Поставь CLI: curl -fsSL https://unyly.org/install | sh
Или настроить вручную
Выполни в терминале:
claude mcp add potluck -- uvx potluckFAQ
Potluck MCP бесплатный?
Да, Potluck MCP бесплатный — установка в пару кликов через Unyly без оплаты.
Нужен ли API-ключ для Potluck?
Нет, Potluck работает без API-ключей и переменных окружения.
Potluck — hosted или self-hosted?
Self-hosted: сервер запускается локально на твоей машине командой из раздела установки.
Как установить Potluck в Claude Desktop, Claude Code или Cursor?
Открой Potluck на unyly.org, выбери вкладку своего клиента (Claude Desktop, Claude Code, Cursor) и нажми Install — конфиг сгенерируется автоматически, без правки JSON.
Похожие MCP
wenb1n-dev/SmartDB_MCP
A universal database MCP server supporting simultaneous connections to multiple databases. It provides tools for database operations, health analysis, SQL optim
автор: wenb1n-devPostgres Server
This server enables interaction with PostgreSQL databases through the Model Context Protocol, optimized for the AWS Bedrock AgentCore Runtime. It provides tools
автор: madhurprashPostgres
Query your database in natural language
автор: AnthropicPostgreSQL
Read-only database access with schema inspection.
автор: modelcontextprotocolCompare Potluck with
Не уверен что выбрать?
Найди свой стек за 60 секунд
Автор?
Embed-бейдж для README
Похожее
Все в категории data
