agentic-compute-mcp

Agentic API Platform: MCP Bridge

A high-performance Model Context Protocol (MCP) server providing premium computational services to autonomous AI agents. This bridge connects Claude and other MCP-compatible LLMs to secure, cloud-hosted endpoints for advanced mathematical optimization and data visualization.

Built with FastMCP and FastAPI, this system leverages the x402 protocol for automated Machine-to-Machine (M2M) microtransactions on the Base network (USDC settlement).

Features

Native MCP Integration: Instantly expose computational tools to Claude Desktop and other MCP clients.

M2M Monetization (x402): Seamless crypto-based settlement per API call. Agents automatically pay for compute using USDC on Base.

Bypass LLM UI Limits: Overcomes LLM token limits by processing massive data structures (like Base64 image matrices) efficiently in the backend.

Available Tools (Endpoints)

This MCP server currently exposes the following premium tools to AI agents:

1. optimize_ga (Cost: 0.50 USDC)

Runs a high-performance Genetic Algorithm (GA) to optimize data models (Polynomial, Logistic, Exponential).

Capabilities: Smart parameter initialization, proportional mutation to prevent premature convergence, and high-accuracy curve fitting (achieves <1% MAPE).

Ideal for: Load forecasting, predictive modeling, and complex hyperparameter tuning.

2. generate_plot (Cost: 0.30 USDC)

A Matplotlib-based rendering engine that generates production-ready charts and graphs.

Capabilities: Bypasses LLM token generation limits by natively drawing data and returning lightweight reference pointers or base64 streams directly to the host machine.

Ideal for: Visualizing optimization results, time-series data, and mathematical models.

Installation & Setup

1. Clone the repository:

git clone https://github.com/codelad1304/agentic-compute-mcp.git cd agentic-compute-mcp

2. Install dependencies:

pip install -r requirements.txt

3. Configure Environment Variables: Create a .env file in the root directory and add your EVM credentials to allow your local MCP client to process agentic payments:

EVM_PRIVATE_KEY=your_private_key

4. Start the MCP Bridge:

python mcp_client.py

Using with Claude Desktop

To install this server for Claude Desktop, add the following to your claude_desktop_config.json:

{ "mcpServers": { "agentic-compute": { "command": "python", "args": ["/absolute/path/to/agentic-compute-mcp/mcp_client.py"] } } }

Author

Kunal Das (Building the financial layer for the agentic web.)

License

This project is licensed under the MIT License - see the LICENSE file for details.