Overview

Dumpling AI MCP Server is a Model Context Protocol (MCP) server implementation that connects MCP-capable clients (like Claude Desktop or Cursor) to Dumpling AI’s suite of data, scraping, processing, agent, and code-execution APIs. It exposes a consistent set of MCP tools that let developers fetch transcripts, crawl and scrape pages, extract structured data, manage knowledge bases, run AI agents, generate images, and execute sandboxed JavaScript/Python — all from within an MCP-enabled environment.

This server is useful when you want to augment an LLM’s context with live web content, structured reverse-engineered extracts, or programmatic computation. It standardizes common tasks (search, scrape, extract, convert, run) into MCP tool calls so a client or agent can orchestrate complex pipelines without writing direct API integrations to Dumpling AI.

Features

Full mapping to Dumpling AI APIs: search, maps, news, YouTube transcripts, autocomplete, and reviews
Web scraping and crawling with JS rendering, content cleaning, screenshots, and structured extraction (AI-driven)
Document conversion and OCR for PDFs, images, audio, and video to text
Knowledge management: ingest scraped/converted content into knowledge bases for agent retrieval
Agent completions: run multi-step AI agents that leverage scraped content and knowledge
Secure developer sandboxes: run JavaScript and Python code safely for data processing
Detailed, structured responses with automatic error handling and standardized MCP formatting
Deployable via Smithery, npx, global npm, and configurable for Cursor

Installation / Configuration

Requirements: a Dumpling AI API key.

Quick install via npx:

# run once with an environment variable
DUMPLING_API_KEY=your_api_key npx -y mcp-server-dumplingai

Install globally:

npm install -g mcp-server-dumplingai
mcp-server-dumplingai   # after installing, set DUMPLING_API_KEY in your environment

Install with Smithery (automates MCP registration for some clients):

npx -y @smithery/cli install @Dumpling-AI/mcp-server-dumplingai --client claude

Cursor configuration example (add under MCP Servers in Cursor settings):

{
  "mcpServers": {
    "dumplingai": {
      "command": "npx",
      "args": ["-y", "mcp-server-dumplingai"],
      "env": {
        "DUMPLING_API_KEY": "<your-api-key>"
      }
    }
  }
}

Common environment variables:

Variable	Required	Description
DUMPLING_API_KEY	Yes	Your Dumpling AI API key used by the server to call Dumpling APIs

Available Tools / Resources

Grouped overview of the built-in MCP tools:

Data APIs
- get-youtube-transcript — extract transcripts with optional timestamps and language
- search — Google-style web search with optional scraping of results
- get-autocomplete — query Google autocomplete suggestions
- search-maps / search-places — location and business searches
- search-news — news search with date ranges
- get-google-reviews — fetch reviews for businesses or places
Web scraping
- scrape — fetch and clean a single page; format as markdown or HTML
- crawl — recursive site crawling with depth/max pages controls
- screenshot — page screenshots with viewport and fullPage options
- extract — AI-guided structured extraction into user-provided schemas
Document conversion
- doc-to-text — PDF/office/image-to-text with OCR options
- media extraction — extract audio/video speech, frames, or metadata
AI & Knowledge
- agent-completion — run agent workflows that leverage knowledge
- knowledge-management — store, query, and update knowledge bases
- image-generation — generate images via Dumpling AI endpoints
Execution tools
- run-javascript — execute JavaScript in a sandbox for small tasks
- run-python — run Python code securely for data processing or formatting

Sample MCP call (scrape):

{
  "name": "scrape",
  "arguments": {
    "url": "https://example.com",
    "format": "markdown",
    "cleaned": true,
    "renderJs": true
  }
}

Sample MCP call (run Python):

{
  "name": "run-python",
  "arguments": {
    "source": "print('hello world')",
    "timeoutMs": 5000
  }
}

Use Cases

Build a knowledge base from a website
- Use crawl to gather pages (crawl -> scrape -> doc-to-text)
- Normalize and extract structured product data with extract
- Ingest results into knowledge-management for agent retrieval
Research + short brief
- run search for query + scrape top N results
- run extract to pull key facts into a simple JSON schema
- call agent-completion to summarize findings and generate recommended next steps
Automated monitoring
- schedule search-news + get-google-reviews for brand mentions
- screenshot pages or capture transcripts for changes
- run a Python snippet to compare current vs previous content and alert on diffs
Data processing and visualization prototyping
- scrape or doc-to-text to get raw text
- run-python to parse, aggregate, and return structured CSV or JSON
- optional: agent uses results to craft a narrative or generate charts
Multimedia transcripts and summarization
- get-youtube-transcript to obtain subtitles or spoken text
- doc-to-text for video captions or audio transcription
- agent-completion produces condensed summaries or action items

Tips & Security

The server requires a valid DUMPLING_API_KEY; keep it secret and rotate when needed.
Execution tools run in confined sandboxes with timeouts — avoid long-running or I/O-heavy jobs.
Use the extract tool to convert messy HTML into reliable structured outputs for downstream agents.
Check responses for standardized error blocks; the server formats errors to help debugging.

Repository and source code: https://github.com/Dumpling-AI/mcp-server-dumplingai

Dumpling AI MCP Server: Scraping, Agents, Execution