CH
Web Scraping

Chrome History AI Chat MCP Server

Chat with an AI about your Chrome browsing history on this MCP server for fun, insights, and personalized highlights.

View on GitHub
Quick Install
npx -y @vincent-pli/chrome-history-mcp

Overview

Chrome History AI Chat MCP Server is a lightweight local server that exposes your Chrome browsing history as structured context for conversational models using the Model Context Protocol (MCP). Instead of handing a raw SQLite file or JSON export to an AI model, this server prepares searchable, chunked, and metadata-rich context so you can ask an LLM questions about what you’ve browsed — timelines, topic clusters, domain summaries, and more.

This setup is useful for developers or privacy-conscious users who want to interact with their own browsing data without uploading it to third-party services. The server runs locally, accepts queries over HTTP, and returns context payloads that an assistant can use to answer prompts, generate summaries, or surface personalized highlights.

GitHub: https://github.com/vincent-pli/chrome-history-mcp

Features

  • Serve Chrome browsing history in an MCP-compatible format for model consumption
  • Load history from Chrome profile SQLite or exported JSON
  • Search and filter history by query, domain, date range, and visit counts
  • Chunk and annotate history entries with metadata (title, URL, visit time, visit type)
  • Simple HTTP API for integrating with local LLMs, LangChain agents, or custom UIs
  • Runs locally — no telemetry or external uploads required

Installation / Configuration

Prerequisites: Node.js (16+), npm or yarn, access to your Chrome profile or an exported history file.

Clone and install:

git clone https://github.com/vincent-pli/chrome-history-mcp.git
cd chrome-history-mcp
npm install
# or
# yarn install

Create a configuration file (.env) from the example and edit values as needed:

cp .env.example .env
# Edit .env to point to your Chrome history file or exported JSON

Typical .env variables (example):

PORT=3000
HISTORY_PATH=/Users/alice/Library/Application Support/Google/Chrome/Default/History
CACHE_ENABLED=true
LOG_LEVEL=info

Run the server in development or production mode:

npm run dev    # watch mode for development
npm start      # production

You should now be able to access the HTTP API at http://localhost:3000. See the API section below for example requests.

Available Resources

  • MCP (Model Context Protocol) — use this server’s endpoints to produce model-ready context blocks
  • Chrome history sources:
    • Chrome SQLite history file (default profile) — often named “History”
    • Exported JSON/CSV exports created by browser extensions or tools
  • Helpful tools:
    • sqlite3 CLI to inspect the History SQLite
    • jq for JSON querying and quick transforms

Common Chrome history file locations:

OSDefault Profile Path
macOS~/Library/Application Support/Google/Chrome/Default/History
Windows%LOCALAPPDATA%\Google\Chrome\User Data\Default\History
Linux~/.config/google-chrome/Default/History

API Example

Search history (example):

curl "http://localhost:3000/search?q=kubernetes&from=2026-01-01&to=2026-03-31&limit=20"

Example response (abbreviated):

{
  "query": "kubernetes",
  "total": 42,
  "items": [
    {
      "id": 123,
      "title": "Kubernetes basics — Example Blog",
      "url": "https://example.com/kubernetes-basics",
      "visitTime": "2026-02-14T10:12:00Z",
      "visitCount": 3,
      "snippet": "An intro to pods, services, and deployments..."
    }
  ]
}

MCP context endpoint (sketch):

curl "http://localhost:3000/mcp/context?session=dev-chat&limit=5"

This returns an MCP-compliant set of context blocks suitable for appending to an assistant prompt.

Use Cases

  • Personal research recall: “What tutorials did I visit about Kubernetes in February 2026?” — pull matching visits, summarize key pages, and create a timeline.
  • Content rediscovery: Ask the assistant to find articles you once opened but didn’t bookmark, e.g., “Find the long read I read about supply chains last year.”
  • Productivity analysis: Summarize frequent domains and top visited pages to reflect where time was spent.
  • Evidence for notes or reports: Collect the sequence of resources consulted for a project and produce a curated reading list with short summaries.
  • Privacy auditing: Inspect third-party domains visited in a given period and export a report for compliance or personal review.

Tips for Developers

  • Run the server locally and point your LLM agent to the MCP endpoints so the model receives context without uploading raw browsing data.
  • Preload or cache parsed history for faster responses if you have a large history file.
  • Use date-range and domain filters to constrain context size and keep model prompts focused.

For full details and advanced configuration, see the repository: https://github.com/vincent-pli/chrome-history-mcp

Tags:web-scraping