CHROME EXTENSION + MCP RUNTIME Live

Facebook Group Search for AI Agents

Search your joined groups, read relevant discussions, and give your AI agent context to work with.

Requires: Signed-in Chrome Facebook session · Local AgentBridge runtime · Compatible MCP client
An AI agent searching selected Facebook groups, reading discussions, and summarizing retrieved findings
Targeted Discovery

Research discussions across the groups you choose.

When evaluating developer tools, market trends, or community feedback, the most practical insights are often scattered across several Facebook communities. Manually searching each group, opening individual posts, and reading comments is slow and tedious.

With Facebook Group Search, you give your agent a specific topic and choose up to 10 joined groups. Your agent coordinates the search in the background, retrieving grounded discussions directly from your authenticated browser session.

SUPPORTED WORKFLOW

Prompt example:

"List my available groups, then search the 3 developer communities I select for discussions about project management tools. Read the top posts and summarize recurring complaints with links."
  • Confined strictly to groups you explicitly select.
  • Filter by recent posts or specific calendar years.
  • Preserves original permalinks when rendered.
MCP Tool Contract

List groups, search posts, then read the discussion.

The capability is broken into three distinct semantic tools exposed to your agent over MCP:

01
facebook_list_groups

Discover & Select

Reads your joined groups list and returns their names, IDs, and links so your agent can ask you which communities to research.

02
facebook_group_search

Search Across Groups

Runs your search across up to 10 selected groups in background tabs. Collects and deduplicates matching posts with recency filters.

03
facebook_group_read_post

Read Full Context

Retrieves the full post body and visible comments for specific posts. When a standalone link is unavailable, it extracts discussion text directly from the rendered card.

Local-First Security

Understand where your research data goes.

Clear boundaries separate your local browser session from downstream model reasoning.

Facebook content passing through a local extension and AgentBridge to an MCP client and its configured model

Loopback-Bound Runtime

The companion @licheesight/agentbridge runtime binds exclusively to 127.0.0.1. It communicates over an authenticated local WebSocket link.

Browser Credentials Stay in Chrome

Your Facebook login cookies, passwords, and session tokens remain inside Chrome. The extension does not export credentials to AgentBridge or write them into configuration files.

Data Flow Follows Your Client

The extension and runtime provide local transport. Once discussion text is retrieved, where it is sent depends on your MCP client and model configuration: local models process data on your machine, while cloud providers receive text per your client's API setup.

Quickstart

Get started with the local setup.

1

Install the Chrome Extension

Install the extension from the Chrome Web Store and keep your Facebook account signed in.

2

Launch the AgentBridge Companion

Run the official local runtime from your terminal:

npx -y @licheesight/agentbridge

Open the printed pairing link (http://127.0.0.1:8787/connect) in Chrome and click Confirm & Connect.

3

Configure Your AI Agent

Generate configuration for Cursor, Claude Code, Windsurf, or generic MCP clients:

npx -y @licheesight/agentbridge setup --client cursor

What the tool can and cannot retrieve

  • Permitted scope: Only searches groups you have already joined and have active permission to view. Does not bypass private group protections.
  • Batch limits: Search queries accept up to 10 explicit group IDs per call to ensure reliable DOM rendering and avoid browser throttling.
  • DOM rendering dependency: Relies on Facebook's rendered web interface. If Facebook omits a permalink or truncates nested comment replies, the extension reports the limitation rather than guessing.
  • Not an automated scraper: Designed for interactive research queries by an AI agent. Does not include 24/7 background monitoring, alerts, or bulk user scraping.
FAQ

Frequently asked questions

Which AI agents and MCP clients can connect?

AgentBridge connects with MCP clients using standard local stdio. The CLI includes setup commands that generate configuration snippets for tools like Cursor, Claude Code, and Windsurf (via npx -y @licheesight/agentbridge setup --client <name>), and works with any environment that can launch a local stdio MCP server process.

Does the extension run while my browser is closed?

No. The extension relies on an active Chrome session to interact with Facebook's rendered DOM. Chrome must be open with your Facebook session signed in.

Does Facebook Group Search perform semantic vector search?

No. The extension executes scoped keyword searches through Facebook's native search interface, then retrieves and ranks rendered posts. Your downstream AI agent performs semantic synthesis and analysis over the retrieved results.

Does LicheeSight store my Facebook data?

No. Neither the Chrome extension nor the local AgentBridge runtime uploads your Facebook posts, cookies, or credentials to any LicheeSight server. All inter-process communication occurs strictly over local loopback (127.0.0.1).

Ready to research your Facebook groups with an AI agent?

Install the Chrome extension and start connecting your agent with AgentBridge.