Project Lou

Lou is an AI-powered assistant/people-search-engine that finds people, writes emails, follows up, and schedules meetings on your behalf—so you can focus on real connections, not repetitive tasks

Comment 2

🧠 About the Project: Project Lou – Your AI-Powered Assistant Suite for everything Outreach via Modular Claude MCPs

🎯 Problem Statement

Despite having state-of-the-art LLMs, we still waste enormous time on repetitive cognitive tasks:

  • Searching for people or companies with vague info
  • Writing personalized cold emails
  • Managing multi-step follow-up logic
  • Scheduling calendar events and parsing replies

I wanted to solve this with a fully modular AI-powered agent that not only “thinks,” but also acts autonomously across the real web and productivity tools.

Because to be honest outreach of many forms job search, sales, recruitment, involves a lot of manual and redundant work that can be automated plus with Claude desktop and MCP it becomes so simple just type in English a small prompt to give the base idea and let prompt engineered anthropic models do the work for you, especially in job search, personalization takes a lot of student’s time and it’s expensive to hire someone or get a service that does it for the student on their behalf, then comes the problem of transparency, sharing of confidential details, up-front costs. Everyone needs it but no one wants to outsource it if they have a better option. With Claude desktop and MCP servers that becomes possible everything remains transparent user has full controls with a super minimalistic UI that when served on dockers can even be made available to mobile devices (imagine just sending a message to Claude desktop in cloud before sleeping and waking up to potentially tens of responses and scheduled meets on calendar just like delegating to a human) by cloud services via a web interface and this can be made commercial while keeping control to the user by charging for the API usage, compute and Claude powered people search engine. I have already received 40+ users in the waitlist I floated within 24hrs, that too passively.

🤖 What is Project Lou?

Project Lou is a suite of specialized Claude-powered MCP servers (Model Context Protocols) that automate outreach workflows — from vague natural language prompts all the way to real email sends and meeting bookings.

Unlike traditional assistants, Lou:

  • Searches the public internet using Google Dorks, not hallucinated guesses
  • Extracts real contact info using APIs
  • Tracks context state across multiple sessions
  • Sends and replies to emails via Gmail
  • Creates, manages, and syncs calendar events

Lou is Claude-powered, but agentic and real-world grounded — it doesn’t just chat, it executes.

🔧 Technical Breakdown

1. 🔍 search_mcp – Prompt-to-Google Dork Translator

  • Converts vague user prompts like:

“Find the guy from the USC SCO club who likes Fight Club and interns at Nvidia”

  • Into structured Google Dork queries using Claude 3 prompt tuning
  • Calls the Google Custom Search API to return real, crawlable links — no hallucination
  • Claude is then used to rank, extract, and explain the results for downstream use

Tech: Claude-3 Opus, Google Custom Search API, prompt chaining, failover keyword expansion

2. 📊 tracker_mcp – URL & Contact Tracker (Firebase)

  • Each link Claude discovers is tracked in Firebase Firestore
  • Links are mapped to one or multiple emails, and a is_new field tracks update status

  • Useful for determining:

    • Who’s been emailed
    • Who replied
    • Who needs a follow-up

  • Exposes two tools: track_url_email, get_tracked_data

Tech: Firebase Admin SDK, Firestore structured schema, custom sync logic with Claude

3. 👤 linkedin-rapidapi-mcp – Structured Profile Enrichment

  • Instead of browser scraping, uses RapidAPI’s LinkedIn Profile endpoint

  • Returns:

    • Full name
    • Role/title
    • Current company
    • Summary

  • Fast, legal, and clean — directly callable by Claude during runtime

Tech: RapidAPI, LinkedIn Profile API, Claude Tool schema

4. 📧 email-finder – Email Discovery MCP (AnyMailFinder)

  • Takes a name + company or LinkedIn URL and returns:

    • Email
    • Validation status
    • Name / Company / Title

  • Fully async, supports bulk lookup via job queuing (prepped for scaling)

Tech: AnyMailFinder API, aiohttp, Claude MCP protocol server

5. 📅 mcp-gsuite – Extended Gmail + Calendar Automation

The original open-source MCP for GSuite had:

  • No email sending
  • Buggy OAuth sync
  • No way to reply, CC, or delete drafts

✅ I patched and extended the server to:

  • Add send_email, reply_to_email, and create_draft tools
  • Fix state desync bugs across .oauth.*.json files
  • Enable complete Google Calendar event creation with metadata (attendees, timezone, reminders)

Now Claude can:

  • Parse replies
  • Schedule meetings
  • Handle real-time inbox ops
  • Add Google Meet links
  • Delete/reschedule/update meetings

Tech: Gmail API, Google Calendar API, OAuth2 flow, tool layering, Claude-compatible action templates

6. 🔎 people-search-mcp – Prompt-based Vague Identity Search Engine

This was one of the most novel additions. I built a Claude-powered identity engine that uses vague inputs like:

“The USC founder who posted on X about LLM scaling laws last week”

And maps them to:

  • Tweets
  • Profiles
  • GitHub repos
  • Email/contact links

It works by chaining:

  • Claude semantic inference
  • Search MCP dork expansion
  • Result parsing
  • Profile cross-matching

This powers true “fuzzy” search – a game-changer for warm outreach.

🧱 Architecture Overview

User Input ─▶ Claude Prompt ─▶ MCP-1 (Search) ─▶ MCP-2 (Link Enrichment)
          │                     └──▶ MCP-3 (Email Discovery)
          └──▶ MCP-4 (Tracker) ─▶ MCP-5 (GSuite) ─▶ Email/Calendar Automation

All servers are local, independently runnable, and Claude Desktop-compatible with tool registration via claude_desktop_config.json.

💥 What Makes It Unique

  • Modular Claude tools — easy to extend or swap components
  • Uses real APIs and data, not hallucinations
  • Full outbound agent — not just question answering
  • Fuzzily finds people with real-world identity clues
  • No scraping, no scraping blockers, no fake proxies

📚 What I Learned

  • How to turn Claude into a full agentic stack, not just a prompt bot
  • Managing sync states between API keys, OAuth, Claude sessions, and Firebase
  • Building real-time search and automation tools from scratch
  • Extending open-source Claude tools responsibly (esp. GSuite)
  • Designing intuitive workflows that actually replace human tasks

🧱 Tech Stack

  • 🧠 Claude 3 Opus
  • 🧰 Python (MCP servers)
  • ☁️ Firebase Firestore
  • 📧 Gmail + Calendar API
  • 🌐 Google Custom Search
  • 🔗 RapidAPI (LinkedIn)
  • 🔒 OAuth2
  • ⚙️ Claude Desktop + MCP spec
  • 🐍 aiohttp, python-dotenv, uvicorn

🛠 Challenges Faced

  • Fixing broken open-source tools (GSuite MCP)
  • Race conditions in .oauth file sync
  • Getting Claude to pass clean structured data between tools (JSON constraints)
  • Handling multiple Claude API response patterns during fallback
  • Coordinating multiple MCPs without central state (stateless Claude call design)

🔓 Open-Sourcing for Builders

This project will be fully open-sourced to help:

  • Students automate job outreach
  • Startups handle cold email at scale
  • Researchers manage knowledge pipelines
  • Anyone tired of repetitive digital grind

You’ll be able to clone individual MCPs or the full stack.

🙌 Final Thought

Project Lou isn’t just about automation — it’s about reclaiming time. It turns Claude into an active partner across your professional workflows, so you can focus on the why, not the how.

Built With

  • admin
  • advanced
  • ai
  • aiohttp
  • anthropic)
  • anymailfinder
  • api
  • beautifulsoup4
  • calendar
  • claude
  • cloud
  • context
  • discovery
  • docker
  • duckduckgo
  • email
  • environment
  • environments
  • fastmcp
  • firebase
  • firebase/firestore
  • git
  • gmail
  • google
  • integration
  • key
  • linkedin
  • management
  • mcp)
  • model
  • oauth2
  • platform
  • powershell
  • protocol
  • python
  • python-dotenv
  • rapidapi)
  • requests
  • sdk
  • search
  • sqlalchemy
  • storage
  • system
  • tracking
  • url/email
  • uvicorn
  • variables
  • venv)
  • virtual
  • windows
  • workspace
Share this project:

Updates