TOOL4LM

All-in-one MCP server for local LLMs (web/doc/scholar/calc) no API keys required.

typescript

GitHub Stars

typescript

Language

6 months ago

First Indexed

2 months ago

Catalog Refreshed

Documentation & install

Readme and setup notes from the catalogue, plus a client-ready config you can copy for your MCP host.

Installation

Add the following to your MCP client configuration file.

Configuration

View docs

{
  "mcpServers": {
    "khanhs-234-tool4lm": {
      "command": "node",
      "args": [
        "--enable-source-maps",
        "C:/path/to/tool4lm/dist/server.js"
      ],
      "env": {
        "SANDBOX_DIR": "C:/path/to/your_docs",
        "ENGINE_ORDER": "searxng,duckduckgo",
        "LANG_DEFAULT": "vi",
        "REGION_DEFAULT": "vn",
        "MAX_FETCH_BYTES": "1048576",
        "FETCH_TIMEOUT_MS": "8000",
        "SEARXNG_ENDPOINTS": "https://searx.be/search,https://searx.tiekoetter.com/search"
      }
    }
  }
}

Tool4LM is a compact MCP server written in Node.js/TypeScript that adds powerful capabilities to local language models. It provides web search, web page reading and summarization, local document access, scholarly metadata lookup, and a calculator, all accessible through a single MCP endpoint to enhance your offline or local LM workflows.

How to use

You run Tool4LM as an MCP server and connect your local LM client to it. The server exposes a set of tools you can invoke from your LM prompts, enabling tasks like web searching, reading and summarizing pages, locating internal documents, retrieving scholarly metadata, and performing calculations. Start by ensuring you can start the local server, then configure your LM client to load Tool4LM as an MCP source. When you ask your model to search the web or read a document, your LM client will route the request to Tool4LM and receive structured results with titles, URLs, summaries, and relevant metadata.

How to install

Prerequisites: Node.js must be installed on your machine. You will also need a working LM client that supports MCP servers (such as LM Studio). Then follow these steps to install and start Tool4LM locally.

Install dependencies

npm install

Build the server

npm run build

Start the server

npm start

When the server starts successfully you should see output like: [TOOL4LM] started. Sandbox: <path_to_sandbox>

Configuration and runtime details

Configure your LM client to load Tool4LM as an MCP source. The server runs locally via Node.js and reads its configuration from an MCP JSON file. The example below shows the minimal setup for a local stdio MCP connection and environment variables that guide its behavior.

{
  "mcpServers": {
    "TOOL4LM": {
      "command": "node",
      "args": ["--enable-source-maps", "C:/path/to/tool4lm/dist/server.js"],
      "type": "mcp",
      "env": {
        "SANDBOX_DIR": "C:/path/to/your_docs",
        "SEARXNG_ENDPOINTS": "https://searx.be/search,https://searx.tiekoetter.com/search",
        "ENGINE_ORDER": "searxng,duckduckgo",
        "LANG_DEFAULT": "vi",
        "REGION_DEFAULT": "vn",
        "MAX_FETCH_BYTES": "1048576",
        "FETCH_TIMEOUT_MS": "8000"
      }
    }
  }
}

Available tools

web_search

Search the web across multiple engines, deduplicate results, and return title, URL, and a snippet.

web_fetch

Fetch a URL with limits on size and time, then extract readable text for summarization.

web_read

Extract readable text from a fetched web page or provided HTML for summarization.

doc_find

Search inside the sandbox directory to locate relevant documents by query.

doc_read

Read a specific file from the sandbox to summarize or quote.

index_build

Build an index of documents to support fast retrieval and summarization.

sch_search

Search scholarly papers and sources across platforms like arXiv, Crossref, and Wikipedia.

sch_get

Retrieve metadata for a paper by DOI, arXivId, or URL.

wiki_search

Search Wikipedia titles by query.

wiki_get

Get Wikipedia summary by title.

calc_eval

Evaluate mathematical expressions with optional precision.

Built by

VeilStrat

AI signals for GTM teams