brenndaude

Bras Droit Platform — AI skill for Claude Code

AI community

AI operations platform for a restaurant: MCP server over Gmail/POS/COGS, Claude vision invoice parsing with a regression harness, tool-using chat agent, Gemini Live voice receptionist, Next.js dashboa.

How to install Bras Droit Platform

This entry records only its repository, not the path inside it, so there is no exact command to give. Open brenndaude/bras-droit-platform and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Bras Droit Platform does

Alternatives in AI

Bot On Anything — A large model-based chatbot builder that can quickly integrate AI models (including ChatGPT, Claude, Gemini) i 4.2k ★
Alibabacloud Ack MCP Server — Alibaba Cloud's ack-mcp-server unifies container operations capabilities, enabling AI assistants and third-par 118 ★
Pingcrm — Personal Networking CRM — AI-powered, open-source, self-hostable 115 ★

README

Bras Droit

Bras Droit ("right-hand man" in French) is an AI operations platform for a small restaurant. It reads the supplier invoices that arrive by email, turns them into structured cost data, ties that data to recipes and point-of-sale sales to compute real margins, and gives the owner and kitchen manager two assistants: a chat agent that can look things up and prepare work, and a voice receptionist that answers the phone.

It was built for, and runs at, ESME, a café-restaurant in Bourg-Saint-Maurice, France. This repository is the code. The client's data is not included; a synthetic demo tenant is provided instead.

What it does

Invoice ingestion. A nightly sweep reads the restaurant's Gmail, finds supplier invoices (PDF, image or email body), parses them into line items and files them against suppliers. Text PDFs go through a local PyMuPDF extractor first; Claude handles the uncertain cases (Haiku on the extracted text, Haiku vision for scans and photos). Every parse is cached by file hash.
MCP server. The ingestion layer is a Python MCP server (esme_mcp, 29 tools) over Gmail, invoices, products, suppliers and the sweep, so the same tools work from Claude Desktop, from scripts and from the API.
Cost and margin model. Invoice lines resolve to canonical ingredients (alias tiers, fuzzy matching, a human review queue for the rest), ingredients feed versioned recipes, recipes feed menu items, and POS exports (Popina) feed sales. The dashboard reports margin leakage, coverage gaps and a daily "can I trust today's numbers" verdict.
Chat agent. A Claude tool-use agent (33 tools in three risk tiers: read, organize, draft) that answers questions from real data and prepares actions such as reorder emails. It never sends anything: drafts go to Gmail for a human to send.
Voice receptionist. A LiveKit worker using Gemini Live. Hours, FAQs, menu document and call routing are edited from the dashboard; the agen