Unified Interface Agents: Establishing Protocol Servers for Data Access

  

Official Documentation Hub   🌪️    Explore the Launch Announcement    🔥    Engage with our Community (Discord)

Model Context Protocol (MCP) specifications empower sophisticated language models, such as Claude, to formulate and execute Hypertext Transfer Protocol (HTTP) requests, thereby bridging the gap to external web services and arbitrary executable scripts. This repository was created both for internal utility and for the broader community of data practitioners seeking a straightforward mechanism to inject arbitrary Python logic directly into their locally deployed Claude instances.

This collection provides a guided, sequential notebook tutorial for provisioning MCP Servers alongside the Claude Desktop Application. The entire system is implemented in Python, built upon Fused's User Defined Function (UDF) architecture.

Prerequisites

• Python interpreter version 3.11 is required.
• The most recent version of the Claude Desktop application must be installed (compatible with macOS and Windows).

Should you operate on a Linux environment where the desktop application is unavailable, we have supplied a lightweight client utility detailed in the section below, allowing local operation.

You are not obligated to possess a Fused subscription for any functionality described herein. All operations are confined and executed exclusively on your local workstation.

Setup Instructions

• Initially, clone this repository into your preferred local directory, and then navigate into the newly created directory:

bash git clone https://github.com/fusedio/fused-mcp.git cd fused-mcp/

• If you lack the package manager uv, please install it:

For macOS / Linux users: bash curl -LsSf https://astral.sh/uv/install.sh | sh

For Windows users:

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

• Verify the installed client functionality by querying its help documentation:

bash uv run main.py -h

• Commence setup by executing our initial guide notebook, fused_mcp_agents.ipynb, within your preferred integrated development environment (IDE). Upon successful configuration, advance to the subsequent notebook to develop custom Agents and functionality interfaces (2.create_your_own_agents.ipynb).

Directory Organization

This repository structure is fundamentally reliant on the MCP Server specification and Fused User Defined Functions (UDFs)—which are standard Python functions capable of remote invocation.

Community Support & Troubleshooting

Should you encounter difficulties during the setup process, we encourage you to seek assistance by joining our Discord server!

Common diagnostic checks include:

• Executing uv run main.py -h should yield output confirming service availability, resembling the following:

• Occasionally, absolute file paths for specific utilities must be explicitly provided within the Claude_Desktop_Config.json file. By default, only uv is referenced generically:

{ "mcpServers": { "qgis": { "command": "uv", "args": ["..."] }

}

}

However, you might need to supply the fully qualified path to the uv executable. This can be easily handled within the notebook using the common.generate_local_mcp_config utility:

python

Within fused_mcp_agents.ipynb

import shutil

common.generate_local_mcp_config( config_path=PATH_TO_CLAUDE_CONFIG, agents_list = ["get_current_time"], repo_path= WORKING_DIR, uv_path=shutil.which('uv'), )

This action results in a configuration file that explicitly defines the path, such as:

{ "mcpServers": { "qgis": { "command": "/Users//.local/bin/uv", "args": ["..."] }

}

}

• If Claude launches but reports no connected tools, consult the official MCP documentation for troubleshooting Claude Desktop integration problems.

Contributions

We welcome external contributions! Feel free to submit Pull Requests to integrate novel UDFs into the udfs/ directory for community exploration.

Utilizing a Local Claude Client (Alternative to Desktop Application)

If installing the Claude Desktop application is infeasible (e.g., on Linux systems), we provide a minimal interface client executable within this project to interact with the locally configured MCP server:

NOTE: This method necessitates an active Anthropic API key, as the Desktop App integration layer is bypassed.

• Register for an account at the Anthropic Console.

• Generate an Anthropic API Key.

• Establish a configuration file named .env: bash touch .env

• Populate the .env file with your key, assigned to the ANTHROPIC_API_KEY variable:

# .env content ANTHROPIC_API_KEY = "your-key-here"

• Launch the MCP server process:

bash uv run main.py --agent get_current_time

• Concurrently, in a separate terminal session, initiate the local client, directing it to the server's endpoint address:

bash uv run client.py http://localhost:8080/sse