openfinch

name	openfinch
description	Self-hosted AI web agent with search, fetch, extract, and browser automation. Use when user asks to research topics, find information, extract data from websites, take screenshots, or automate web browsing tasks.
version	0.1.0
author	OpenFinch
homepage	https://github.com/JonusNattapong/OpenFinch
triggers	["research","search the web","find information","extract data from","scrape","take a screenshot of","browse to","go to","navigate to","extract pricing","extract product","extract information from","summarize the page","automate web","web agent"]

OpenFinch — Web Agent Skill

OpenFinch is a self-hosted AI web agent infrastructure available as an MCP server. Use these tools when the task involves web research, data extraction, or browser automation.

Available Tools

Tool	Use For
openfinch_search	Web search when you don't have a URL
openfinch_fetch	Get page content as markdown
openfinch_extract	Structured data extraction with LLM
openfinch_browser_create_session	Start browser for JS-heavy pages
openfinch_browser_screenshot	Take screenshot
openfinch_browser_close_session	Clean up browser session
openfinch_agent_run	Autonomous agent for multi-step tasks
openfinch_agent_get_result	Get agent result
openfinch_agent_get_events	Get agent event trace

Quick Start

Before using, ensure OpenFinch is running:

# One-line deploy
curl -fsSL https://raw.githubusercontent.com/JonusNattapong/OpenFinch/main/deploy.sh | bash

# Or manual
git clone https://github.com/JonusNattapong/OpenFinch.git
cd OpenFinch && cp .env.example .env
# Edit .env and set LLM keys
docker compose up -d

Configure in Claude Code (~/.claude/settings.local.json):

{
  "mcpServers": {
    "openfinch": {
      "command": "npx",
      "args": ["-y", "@openfinch/mcp-server"],
      "env": {
        "OPENFINCH_API_URL": "http://localhost:8787"
      }
    }
  }
}

Usage Patterns

1. Simple Search

Search for information: use openfinch_search with query

2. Fetch Page Content

Get page content: use openfinch_fetch with url and format="markdown"

3. Extract Structured Data

Extract data: use openfinch_extract with url and prompt describing what to extract

4. Browser Automation

1. Create session: openfinch_browser_create_session
2. Navigate: use openfinch_fetch with the URL (or agent_run for complex tasks)
3. Screenshot: openfinch_browser_screenshot with sessionId
4. Close: openfinch_browser_close_session

5. Multi-Step Agent Task

1. Start agent: openfinch_agent_run with goal (and optional startUrl)
2. Poll for result: openfinch_agent_get_result with runId
3. Get full trace: openfinch_agent_get_events with runId

Safety

• Destructive actions blocked: delete, purchase, submit payment, send messages require explicit user confirmation
• Allowed domains: respect allowedDomains if specified
• robots.txt: respected by default
• No credentials: never enter passwords or secrets unless user explicitly provides them

Examples

Research a topic: → openfinch_search → openfinch_fetch → openfinch_extract

Get structured data: → openfinch_extract with prompt describing the schema

Multi-page research: → openfinch_agent_run with goal describing the full task

Screenshot a page: → openfinch_browser_create_session → openfinch_browser_screenshot → openfinch_browser_close_session