# Desearch [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/Desearch-ai/mcp-desearch  
**GitHub Stars:** 1  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/desearch

## Description
AI search, X search, web search, page extraction, and X trends.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "desearch": {
    "command": "npx",
    "args": ["-y","desearch-mcp-server"],
    "env": {
      "DESEARCH_API_KEY": ""
    }
  }
}
```

**Requires environment variables:** `DESEARCH_API_KEY` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What Desearch MCP server does

Desearch MCP server connects MCP-compatible agents to Desearch search and X data APIs. It covers AI-assisted searches across web and Twitter, SERP-style web searches, link discovery, public page extraction, and a broad set of X post, user, reply, retweeter, and trend queries.

The server requires a Desearch API key. It can run as a local stdio process installed from npm, or clients can connect to the hosted endpoint at `https://mcp.desearch.ai/mcp`. The local package requires Node.js 20.18.1 or later; Node.js 18 is not supported.

## How it works

For local use, install the `desearch-mcp-server` npm package globally or launch it with `npx`. Configure the MCP client to start the package and provide `DESEARCH_API_KEY` in its environment. The hosted option uses the same MCP interface but passes the key through an `x-api-key` header.

AI Search accepts source identifiers such as `web`, `twitter`, `arxiv`, `wikipedia`, `hackernews`, and `reddit`; web and Twitter are selected by default. Older source labels are rewritten to the supported short identifiers. The API does not accept `youtube` as an AI Search source.

Page reading is handled by `extract`, which returns a public URL as text or HTML and can accept JavaScript execution and a wait time. `web-crawl` remains available for the legacy crawl route, but new integrations should prefer `extract`.

## Setup and configuration

Create a Desearch API key through the Desearch API console before configuring the server. A local stdio configuration uses the following environment variable:

```json
{
  "mcpServers": {
    "desearch": {
      "command": "npx",
      "args": ["-y", "desearch-mcp-server"],
      "env": { "DESEARCH_API_KEY": "your-api-key" }
    }
  }
}
```

The repository documents setup for Claude Desktop, Cursor, Windsurf, Zed, Gemini CLI, and Cline. Gemini CLI installs an extension from the repository and uses the hosted endpoint. Windsurf and Zed also support hosted configurations with the `x-api-key` header. Smithery installation is documented for Claude Desktop and Cursor.

## Tools and capabilities

Available tools include:

- `ai-search` for summarized AI searches with relevant links.
- `x-search` for filtered X tweet searches, including author, date, language, verification, media, engagement, and quote filters.
- `web-search` and `web-links-search` for web results and link-oriented searches.
- `extract` and the deprecated-compatible `web-crawl` for reading public URLs.
- `x-links-search` for finding X post links through AI search.
- `x-posts-by-urls` and `x-post-by-id` for retrieving specific posts.
- `x-posts-by-user`, `x-user-posts`, and `x-user-replies` for user activity.
- `x-post-retweeters` and `x-post-replies` for post relationships.
- `x-trends` for location-based trending topics using a WOEID.

## Limitations and notes

The Desearch MCP server does not expose the removed `latestTweets` SDK method or its former `/twitter/latest` endpoint. `web-links-search` only accepts the `web` source, while `x-trends` requires a WOEID and supports counts from 30 to 100. Search and extraction behavior depends on the Desearch API and requires a valid key.

_Full upstream README: https://allmcps.com/mcp/desearch/readme_