morph-websearch-mcp

MCP server that brings web search, page fetching, and AI-powered research to any MCP-compatible client. Backed by crawl4ai for crawling and morphllm for context compaction and agent reasoning.

Why

LLMs drown in boilerplate. Every webpage you feed them includes nav bars, cookie banners, tracking hashes, and ads. morph-websearch-mcp solves this by compacting every page through morph compact — a 33,000 tok/s engine that drops irrelevant lines without rewriting — before the content ever reaches your LLM.

Related MCP server: DuckDuckGo MCP Server

Tools

Install

pip install morph-websearch-mcp
# or
uv add morph-websearch-mcp

Setup

Get an API key from morphllm, then:

export MORPH_API_KEY="sk-..."

MCP Client Config

{
  "mcpServers": {
    "websearch": {
      "command": "websearch-mcp",
      "env": { "MORPH_API_KEY": "sk-..." }
    }
  }
}

How it works

query → DuckDuckGo HTML → crawl4ai (parallel fetch) → morph compact → clean results
                                                                          ↓
                                                              morph-dsv4flash agent loop
                                                        (only for webextract — searches
                                                         and fetches autonomously)

Every page fetched by websearch and webfetch passes through morph compact before being returned. The webextract agent uses morph-dsv4flash (DeepSeek V4 Flash, ~150 tok/s) in an OpenAI-compatible tool-calling loop to find answers across multiple search/fetch cycles.

Services used

License

MIT