MCP Support¶

Complete guide to Model Context Protocol (MCP) integration with ToolUniverse

ToolUniverse provides comprehensive support for the Model Context Protocol (MCP), enabling seamless integration with AI scientists, reasoning models, and agentic systems. This guide covers everything you need to know about using ToolUniverse through MCP.

What is MCP?¶

The Model Context Protocol (MCP) is a standardized protocol that enables AI scientists to securely connect to external tools and data sources. ToolUniverse implements MCP through the Scientific Model Context Protocol (SMCP), extending standard MCP capabilities with scientific domain expertise.

Key Benefits: - Standardized Integration: Connect to any MCP-compatible AI scientist - Scientific Tool Access: Direct access to 649+ scientific tools - Intelligent Discovery: AI-powered tool search and recommendation - Secure Communication: Standardized protocol ensures secure tool execution - Production Ready: High-performance architecture for real-world applications

MCP Architecture Overview¶

AI Scientist (Claude, ChatGPT, Gemini, etc.)
        │
        │ MCP Protocol
        │
┌─────────────────┐
│ ToolUniverse    │ ← MCP Server
│   MCP Server    │
└─────────────────┘
        │
        │ Tool Execution
        │
┌─────────────────┐
│ Scientific      │
│ Tools (649+)    │
└─────────────────┘

ToolUniverse MCP Implementation¶

ToolUniverse provides three main MCP server implementations:

`tooluniverse-smcp` - Full-featured server with configurable transport (HTTP, SSE, stdio)
`tooluniverse-smcp-stdio` - Specialized server for stdio transport (optimized for desktop AI applications)

All servers expose the same comprehensive set of 649+ scientific tools through the MCP protocol.

Quick Start¶

For basic MCP server setup and configuration, see the comprehensive guide in MCP Server Functions.

CLI Options Reference¶

The following are commonly used command-line flags for ToolUniverse MCP servers.

tooluniverse-smcp [OPTIONS]

--port INT                     Server port (HTTP/SSE). Default: 7000
--host TEXT                    Bind host for HTTP/SSE. Default: 0.0.0.0
--transport [http|stdio|sse]   Transport protocol. Default: http
--name TEXT                    Server display name
--max-workers INT              Worker pool size for tool execution
--verbose                      Enable verbose logs

# Tool selection
--categories STR...            Include only these categories
--exclude-categories STR...    Exclude these categories
--include-tools STR...         Include only these tool names
--tools-file PATH              File with one tool name per line
--include-tool-types STR...    Include only these tool types
--exclude-tool-types STR...    Exclude these tool types
--tool-config-files TEXT       Mapping like "custom:/path/to/custom.json"

# Hooks
--hooks-enabled                Enable hooks (default: False)
--hook-type [SummarizationHook|FileSaveHook]
--hook-config-file PATH        JSON config for hooks

tooluniverse-smcp-stdio [OPTIONS]

--name TEXT                    Server display name
--categories STR...            Include only these categories
--include-tools STR...         Include only these tool names
--tools-file PATH              File with one tool name per line
--include-tool-types STR...    Include only these tool types
--exclude-tool-types STR...    Exclude these tool types
--hooks                        Enable hooks (default: disabled for stdio)
--hook-type [SummarizationHook|FileSaveHook]
--hook-config-file PATH        JSON config for hooks

Configuration¶

All MCP servers support configuration through command-line arguments. See the CLI Options Reference above for available configuration options.

Configuration Files¶

Example tools file (one tool per line, lines starting with # are comments):

# tools.txt
OpenTargets_get_associated_targets_by_disease_efoId
Tool_Finder_LLM
ChEMBL_search_similar_molecules
# Tool_Finder_Keyword

Example hook config file:

{
  "SummarizationHook": {
    "max_tokens": 2048,
    "summary_style": "concise"
  },
  "FileSaveHook": {
    "output_dir": "/tmp/tu_outputs",
    "filename_template": "{tool}_{timestamp}.json"
  }
}

Client Integration Examples¶

Python MCP client (conceptual) connecting to HTTP server:

import requests

# Discover tools
tools = requests.get("http://127.0.0.1:8000/mcp/tools").json()

# Execute a tool
payload = {
    "name": "UniProt_get_entry_by_accession",
    "arguments": {"accession": "P04637"}
}
result = requests.post("http://127.0.0.1:8000/mcp/run", json=payload).json()
print(result)

JavaScript MCP client (conceptual) against HTTP server:

  const fetch = require('node-fetch');

  async function run() {
    const toolsResp = await fetch('http://127.0.0.1:8000/mcp/tools');
    const tools = await toolsResp.json();
    console.log('Tools:', tools.length);

    const resp = await fetch('http://127.0.0.1:8000/mcp/run', {
      method: 'POST',
      headers: { 'Content-Type': 'application/json' },
      body: JSON.stringify({
        name: 'UniProt_get_entry_by_accession',
        arguments: { accession: 'P04637' }
      })
    });
  const result = await resp.json();
  console.log(result);
}
 run();

Streaming Output¶

All MCP-exposed tools now accept an optional _tooluniverse_stream flag. When set to true, compatible tools send incremental text chunks as MCP log notifications while still returning the final result payload at completion. Example request payload:

{
  "method": "tools/call",
  "params": {
    "name": "AgenticTool_example",
    "arguments": {
      "question": "Summarise recent literature",
      "_tooluniverse_stream": true
    }
  }
}

Make sure your client surfaces notifications/log (FastMCP ctx.info) messages to display the streamed output.

Claude Desktop stdio registration (example):

{
  "mcpServers": {
    "tooluniverse": {
      "command": "tooluniverse-smcp-stdio",
      "args": ["--categories", "uniprot", "ChEMBL", "opentarget", "--hooks", "--hook-type", "SummarizationHook"]
    }
  }
}

MCP Server Configuration¶

Transport Options¶

ToolUniverse MCP servers support multiple transport protocols:

HTTP Transport (Default)

Best for web-based applications and remote access
Supports RESTful API endpoints
Configurable host and port

STDIO Transport

Optimized for desktop AI applications
Direct process communication
Lower latency for local applications

Server-Sent Events (SSE)

Real-time streaming capabilities
Suitable for interactive applications
Supports long-running operations

Tool Selection¶

Configure which tools are available through the MCP server. For detailed configuration options including category-based loading, tool-specific loading, and type-based filtering, see Category-Based Loading, Tool-Specific Loading, and Type-Based Filtering.

Advanced Configuration¶

Hook Configuration¶

Enable intelligent output processing hooks for MCP servers. For comprehensive hook configuration including SummarizationHook and FileSaveHook, see Hook Configuration.

See also

Detailed Guide: Server and Stdio Hook Integration - Complete hook integration tutorial

Performance Tuning¶

Optimize server performance for your use case. For detailed performance configuration options, see Server Configuration.

AI Scientist Integration¶

ToolUniverse MCP servers are compatible with major AI scientists and platforms:

Claude Desktop¶

Integrate ToolUniverse with Claude Desktop for powerful desktop-based scientific research.

See also

For complete Claude Desktop integration, see Claude Desktop

Tutorial: ToolUniverse MCP Integration with Claude Desktop App - Step-by-step Claude Desktop setup

ChatGPT API¶

Connect ToolUniverse to ChatGPT API for programmatic AI-scientist workflows.

See also

For ChatGPT API integration, see ChatGPT API

Gemini CLI¶

Use ToolUniverse with Gemini CLI for command-line scientific research.

See also

For Gemini CLI integration, see Gemini CLI

Tutorial: ToolUniverse MCP Integration with Gemini CLI - Complete Gemini CLI setup guide

Claude Code¶

Integrate ToolUniverse with Claude Code for IDE-based scientific development.

See also

For Claude Code integration, see Claude Code

Qwen Code¶

Connect ToolUniverse to Qwen Code for terminal-based scientific workflows.

See also

For Qwen Code integration, see Qwen Code

GPT Codex CLI¶

Use ToolUniverse with GPT Codex CLI for advanced command-line research capabilities.

See also

For GPT Codex CLI integration, see GPT Codex CLI

MCP Protocol Details¶

Tool Discovery¶

MCP clients can discover available tools through the standard MCP protocol. For detailed tool discovery methods and examples, see mcp-server-integration.

Tool Execution¶

Execute tools through the MCP protocol. For comprehensive tool execution patterns and MCP client examples, see mcp-client-integration.

Error Handling¶

MCP provides standardized error handling. For detailed error handling patterns and troubleshooting, see error-handling-validation.

MCP Server Management¶

Server Status¶

Monitor MCP server status and health. For server management commands and status monitoring, see Discovery Commands.

Logging and Debugging¶

Enable comprehensive logging for debugging. For detailed logging configuration and debugging options, see Method 2: Global Configuration.

Performance Monitoring¶

Monitor MCP server performance. For performance monitoring and optimization, see performance-optimization.

Troubleshooting¶

Common Issues¶

MCP Server Not Starting

Check if port is available
Verify ToolUniverse installation
Check server logs for error messages

Tools Not Available

Verify tool categories are loaded
Check tool names are correct
Ensure tools are not excluded

Connection Issues

Verify transport protocol matches client expectations
Check firewall settings for HTTP transport
Ensure proper authentication for remote connections

Performance Issues

Increase worker threads
Enable caching for repeated tool calls
Use specific tool categories instead of loading all tools

For comprehensive troubleshooting guide, see troubleshooting.

Debug Commands¶

Useful debugging commands and validation methods. For complete debugging command reference, see Discovery Commands.

Best Practices¶

Security¶

Use HTTPS in production environments
Implement proper authentication and authorization
Regularly update ToolUniverse and MCP dependencies
Monitor server logs for suspicious activity

Performance¶

Load only necessary tool categories
Use appropriate worker thread counts
Enable caching for frequently used tools
Monitor server metrics and adjust configuration

Reliability¶

Implement proper error handling in MCP clients
Use retry mechanisms for transient failures
Monitor server health and restart if needed
Keep backup configurations for critical deployments

For detailed best practices and production deployment guidance, see performance-optimization.

Summary¶

ToolUniverse’s MCP support provides a powerful, standardized way to integrate scientific tools with AI scientists. The SMCP implementation extends standard MCP capabilities with scientific domain expertise, making it easy to build sophisticated AI-scientist workflows.

Key takeaways:

Easy Integration: Simple setup with major AI scientists
Comprehensive Tools: Access to 649+ scientific tools through MCP
Flexible Configuration: Multiple transport options and tool selection
Production Ready: High-performance, secure, and reliable
Extensive Documentation: Complete guides for all major AI platforms

Start with the Building AI Scientists guide to begin building your AI scientist, or explore specific integrations for your preferred AI scientist.

MCP Support¶

What is MCP?¶

MCP Architecture Overview¶

ToolUniverse MCP Implementation¶

Quick Start¶

CLI Options Reference¶

Configuration¶

Configuration Files¶

Client Integration Examples¶

Streaming Output¶

MCP Server Configuration¶

Transport Options¶

Tool Selection¶

Advanced Configuration¶

Hook Configuration¶

Performance Tuning¶

AI Scientist Integration¶

Claude Desktop¶

ChatGPT API¶

Gemini CLI¶

Claude Code¶

Qwen Code¶

GPT Codex CLI¶

MCP Protocol Details¶

Tool Discovery¶

Tool Execution¶

Error Handling¶

MCP Server Management¶

Server Status¶

Logging and Debugging¶

Performance Monitoring¶

Troubleshooting¶

Common Issues¶

Debug Commands¶

Best Practices¶

Security¶

Performance¶

Reliability¶

Related Documentation¶

Core MCP Components¶

AI Scientist Integration¶

MCP Tutorials and Guides¶

Advanced Features¶

MCP Tools and Examples¶

Examples and Tutorials¶

API Reference¶

External Resources¶

Summary¶