home / mcp / tooluniverse mcp server

ToolUniverse MCP Server

Provides an MCP server for ToolUniverse enabling remote and local tool access for AI scientist workflows.

Installation

Add the following to your MCP client configuration file.

Configuration

{
  "mcpServers": {
    "mims-harvard-tooluniverse": {
      "command": "tooluniverse-smcp",
      "args": []
    }
  }
}

You can run and connect to ToolUniverse as an MCP server to enable programmable, pipeline-driven AI science workflows. This MCP server exposes a controlled interface that lets you load a vast ecosystem of tools, discover relevant endpoints, and execute them through a unified protocol, enabling AI models to reason, design experiments, and analyze results using standardized tool calls.

How to use

To use ToolUniverse as an MCP server, start the local MCP server binary and connect with an MCP client. Once running, you can load the available tools, search for tools by description or keyword, and execute selected tools to perform data analysis, retrieval, or experimental tasks.

Key workflows you can perform include loading the tool ecosystem, discovering tools that match your scientific query, and calling those tools to obtain results. You can chain tools into simple workflows or multi-step pipelines, enabling sequential or parallel execution under a self-directed discovery process.

How to install

Prerequisites: you should have Python and a compatible runtime environment installed on your system.

Step 1: Install the ToolUniverse package from PyPI.

pip install tooluniverse

Step 2 (optional): If you intend to expose the MCP server over HTTP for remote clients, install the HTTP API component and start the server with the desired host and port.

pip install tooluniverse
tooluniverse-http-api --host 0.0.0.0 --port 8080

Step 3: Start the MCP server locally using the standard MCP server command.

tooluniverse-smcp

Step 4 (optional): If you plan to connect Claude or other MCPB-enabled clients, configure Claude to use the stdio MCP transport for ToolUniverse as shown.

pip install tooluniverse
claude mcp add --transport stdio tooluniverse -- tooluniverse-smcp-stdio --compact-mode

Additional notes

The MCP server exposes a robust tool ecosystem that supports loading 1000+ scientific tools and provides a standardized interface for tool requests and results. You can discover tools using a keyword-based finder and then call a specific tool to obtain results. You can also compose tools into workflows to run sequences or parallel steps.

If you want to expose the server remotely, you can use the HTTP API option described above, which allows remote clients to load tools, discover endpoints, and call tools with minimal client dependencies.

Available tools

Tool_Finder_Keyword

Finds relevant tools by embedding or keyword description to match a research task.

OpenTargets_get_associated_targets_by_disease_efoId

Retrieves targets associated with a disease using an EFO identifier (example: hypertension).

UniProt_get_function_by_accession

Fetches the function and annotations for a protein by its UniProt accession.

ToolUniverse_load_tools

Loads the full set of available tools into the MCP runtime for discovery and use.