Skip to content
MCP server

Wet

Self-hosted MCP server for web search, page extraction and library docs

Install
Pricing
Open source
Platforms
self-hosted, CLI
Runs
Remotely
Version
3.22.0
Listing updated
today

Similar MCP servers

About Wet

Wet is an open-source MCP server for web search and scraping. It runs as a hosted remote server, so there is nothing to install locally.

Wet runs as a local HTTP service rather than a subprocess your client spawns. You start it with wet server start, then point Claude Code, Codex, Gemini CLI, Cursor or Windsurf at http://127.0.0.1:8000/mcp; a Docker compose stack with token auth is offered for always-on setups.

Search goes through a fallback chain of backends (Tavily and DuckDuckGo by default, with Brave, Exa, Kagi, OpenRouter, Firecrawl and Startpage selectable) and results are cached for an hour, or five minutes for time-sensitive queries. A separate academic mode covers Google Scholar, Semantic Scholar, arXiv, PubMed, CrossRef and BASE.

Page extraction escalates from plain HTTP through TLS spoofing to browser rendering and an optional captcha step, and returns chunked text, markdown, JSON-LD and code blocks. It can batch up to 50 URLs, crawl sites and convert local PDF, DOCX, XLSX, CSV, HTML, EPUB and PPTX files to markdown. Embedding and reranking run locally by default, and an OpenRouter key switches them to cloud models.

Install

Run in a terminal

claude mcp add wet --env OPENROUTER_API_KEY='<OPENROUTER_API_KEY>' --env GITHUB_TOKEN='<GITHUB_TOKEN>' -- uvx n24q02m-wet

Add to claude_desktop_config.json

{
  "mcpServers": {
    "wet": {
      "command": "uvx",
      "args": [
        "n24q02m-wet"
      ],
      "env": {
        "OPENROUTER_API_KEY": "<OPENROUTER_API_KEY>",
        "GITHUB_TOKEN": "<GITHUB_TOKEN>"
      }
    }
  }
}

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "wet": {
      "command": "uvx",
      "args": [
        "n24q02m-wet"
      ],
      "env": {
        "OPENROUTER_API_KEY": "<OPENROUTER_API_KEY>",
        "GITHUB_TOKEN": "<GITHUB_TOKEN>"
      }
    }
  }
}

Add to .vscode/mcp.json

{
  "servers": {
    "wet": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "n24q02m-wet"
      ],
      "env": {
        "OPENROUTER_API_KEY": "<OPENROUTER_API_KEY>",
        "GITHUB_TOKEN": "<GITHUB_TOKEN>"
      }
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "wet": {
      "command": "uvx",
      "args": [
        "n24q02m-wet"
      ],
      "env": {
        "OPENROUTER_API_KEY": "<OPENROUTER_API_KEY>",
        "GITHUB_TOKEN": "<GITHUB_TOKEN>"
      }
    }
  }
}

Add to your client's MCP settings (most use this shape)

{
  "mcpServers": {
    "wet": {
      "command": "uvx",
      "args": [
        "n24q02m-wet"
      ],
      "env": {
        "OPENROUTER_API_KEY": "<OPENROUTER_API_KEY>",
        "GITHUB_TOKEN": "<GITHUB_TOKEN>"
      }
    }
  }
}

Using several servers? Build one config file with the MCP config generator.

Configuration

  • OPENROUTER_API_KEYOpenRouter API key — the single default credential for the [models.embed|rerank|chat|jev_score] provider cells and the openrouter search backend. Per-task overrides live in ~/.wet/config.toml; keys may also be injected via HULL_EMBED_API_KEY / HULL_RERANK_API_KEY / HULL_CHAT_API_KEY / HULL_JEV_SCORE_API_KEY. Empty cells fall back to local ONNX models.
  • GITHUB_TOKENGitHub personal access token for higher rate limits on library discovery

What it does

Multi-backend web search
Falls back across several search providers with query expansion and a short-lived cache.
Academic paper search
Queries Google Scholar, Semantic Scholar, arXiv, PubMed, CrossRef and BASE.
Library docs index
Discovers and indexes documentation with hybrid full-text search and version-specific lookups.
Escalating page extraction
Moves from plain HTTP to browser rendering until a page can be read, with batch and crawl modes.
Local file conversion
Turns PDF, DOCX, XLSX, CSV, HTML, EPUB and PPTX files into markdown.
Local models by default
Embeds and reranks without API keys; OpenRouter is optional.

Good for

  • Give a coding agent version-aware library documentation
  • Collect and cite sources for a research question across web and academic indexes
  • Extract clean text and structured data from a batch of URLs
  • Convert a folder of office documents to markdown before analysis

Pricing

Open source

Open-source; optional third-party API keys for search and models

Wet FAQ

Does Wet run over stdio?

No. The README says it is HTTP-only: you start a local instance first and register its streamable-HTTP URL with your client.

Which package do I install?

The PyPI distribution is n24q02m-wet; `uvx --python 3.13 --from n24q02m-wet wet server start` starts it.

Do I need API keys?

Not for the defaults. Tavily, GitHub and OpenRouter keys are optional and raise quality or rate limits.

What happened to image analysis?

The media analyze action was removed in v2.0.0; the README points to the separate imagine-mcp project for vision and audio.

Is Wet free?

Yes. Wet is open source, so you can use and self-host it for free.

How do I add Wet to Claude Code?

Run this command in your terminal: claude mcp add wet --env OPENROUTER_API_KEY='<OPENROUTER_API_KEY>' --env GITHUB_TOKEN='<GITHUB_TOKEN>' -- uvx n24q02m-wet

What are the best alternatives to Wet?

The closest alternatives to Wet are Scrapling MCP Server, Firecrawl MCP Server, Douyin_TikTok_Download_API, SearXNG Search and machine_internet.

Where this comes from

  • Description: written from the project's README.
  • Install steps: from the official MCP Registry entry.

How AIPoweredBot checks servers