beeper-chat-summarizer

Beeper AI Chat Summarizer (Local MCP Edition)

This project transforms your local machine into a private AI assistant for your Beeper chats. Built on top of the Self-Hosted AI Starter Kit, it utilizes Beeper's local Model Context Protocol (MCP) to securely stream chat messages into a local Postgres database. When a user types /summarize, an n8n automation pipeline fetches the recent chat history, processes it through a local LLM (Ollama / Llama 3.2), and beams a formatted summary directly back into the chat.

100% self-hosted. Zero cloud APIs. Absolute privacy.

Key Features

• Total Privacy: Uses Beeper's local Developer Bridge and Ollama. Your chat logs never touch OpenAI, Google, or any external cloud.
• On-Demand Summaries: Just type /summarize in your selected Beeper chat to trigger the AI pipeline.
• Smart Caching: Maintains a persistent Postgres connection to log messages efficiently and avoid duplicate summaries.
• Advanced Networking Solutions: Includes built-in Windows network proxies and registry tweaks to solve common Docker-to-Host communication blocks and socket exhaustion (WinError 10055).

---

Architecture

• Beeper Local MCP Server: Enabled via Beeper Developer Options. Exposes a local WebSocket and HTTP API for direct, standard-compliant access to chat streams.
• Self-Hosted AI Starter Kit: The Dockerized foundation providing our database, automation engine, and LLM.
• Python Collector (main.py): Runs natively on Windows. Listens to the Beeper MCP WebSocket, maintains a persistent Postgres connection, and triggers the n8n webhook via API calls.
• Postgres (Docker): Stores the chat history and prevents duplicate messages.
• n8n (Docker): The automation engine. Grabs chat logs, sends them to Ollama, and POSTs the summary back to Beeper.
• Ollama (Docker): Runs the local LLM (llama3.2:latest) for sentiment analysis and summarization.

---

Prerequisites

• Windows OS (Instructions heavily feature Windows networking/PowerShell).
• Docker Desktop installed and running.
• The Self-Hosted AI Starter Kit cloned and running (docker-compose up -d).
• Beeper Desktop installed with Developer Options / Local Bridge enabled (Find your Port and Token in Beeper Settings).
• Python 3.x installed.

---

Step 1: Python MCP Collector Setup

The Python script (main.py included in this repo) acts as the middleman between the Beeper MCP server and your Starter Kit. It uses websockets, psycopg2, and requests to listen for messages.

It specifically maintains a persistent database connection to avoid Windows socket exhaustion and ignores the AI's own summary messages (HTML tags) so they don't loop endlessly.

1. Install dependencies: pip install websockets psycopg2 requests

2. Configure main.py: Open the script and update the configuration variables at the top of the file with your specific credentials:

• BEEPER_WS (Ensure the port matches your Beeper settings, e.g., 23375)
• BEEPER_TOKEN
• ALLOWED_CHATS
• IGNORED_SENDERS

3. Run the script: python main.py

---

Step 2: n8n Workflow Configuration

Instead of building the automation from scratch, you can simply import the included workflow into your AI Starter Kit n8n instance!

1. Open your local n8n instance (http://localhost:5678).
2. Go to Workflows -> Add Workflow.
3. Click the ... (Options) menu in the top right and select Import from File.
4. Upload the workflow.json file provided in this repository.
5. Update your credentials for Postgres and Ollama inside the nodes.
6. Toggle the workflow to Publish/Active.

Critical Beeper MCP HTTP Request Node Note

Beeper's local API rejects requests originating from Docker containers. The included workflow bypasses this using a Host: localhost header spoof. If your Beeper MCP changes ports (e.g., from 23373 to 23375), make sure to update the URL in the final HTTP Request node.

---

Step 3: Windows Networking & Docker Fixes

Because n8n runs inside Docker and the Beeper MCP server runs on the Windows host, Docker needs permission to talk to the host machine's localhost.

1. Create a Port Proxy (Run in PowerShell as Administrator): This forces Windows to route Docker traffic to the local Beeper API. netsh interface portproxy add v4tov4 listenport=23375 listenaddress=0.0.0.0 connectport=23375 connectaddress=127.0.0.1

2. Open the Firewall: New-NetFirewallRule -DisplayName "Allow Beeper for n8n" -Direction Inbound -LocalPort 23375 -Protocol TCP -Action Allow

---

Step 4: Fixing Windows Socket Exhaustion (WinError 10055)

If you experience network crashes or WinError 10055, Windows has likely run out of socket buffer space due to holding old connections to the database and Beeper in a TIME_WAIT state for too long.

Run these registry tweaks in PowerShell as Administrator to fix socket exhaustion permanently:

# Reduce TIME_WAIT duration from 4 minutes to 30 seconds
Set-ItemProperty -Path "HKLM:\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters" -Name "TcpTimedWaitDelay" -Value 30 -Type DWord

# Increase the port range so Windows has more ports to work with
Set-ItemProperty -Path "HKLM:\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters" -Name "MaxUserPort" -Value 65534 -Type DWord

# Allow port reuse faster
Set-ItemProperty -Path "HKLM:\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters" -Name "StrictPortReuse" -Value 1 -Type DWord