EIPs MCP

An MCP server for providing semantically related Ethereum Improvement Proposals (EIPs) to AI agents.

Features

• Semantic Search: Utilizes vector embeddings for accurate, context-aware search across EIP documents.
• Markdown Support: Processes EIP markdown files with chunking for efficient storage and retrieval.

Installation

Prerequisites

• Python 3.10+
• uv (recommended) or pip for dependency management

Setup

1. Clone the Repository:

git clone --recursive-submodules https://github.com/kukapay/eips-mcp.git
   cd eips-mcp

2. Install the Dependencies (recommended):

uv sync

3. Load EIP Documents:

Use the provided load_eips.py script to automatically fetch and process EIPs:

uv run load_eips.py

4. Installing to Claude Desktop:

Install the server as a Claude Desktop application:

uv run mcp install main.py --name "EIPs"

Configuration file as a reference:

{
       "mcpServers": {
           "EIPs": {
               "command": "uv",
               "args": [ "--directory", "/path/to/eips-mcp", "run", "main.py" ]
           }
       }
    }

Replace /path/to/eips-mcp with your actual installation path.

Usage

The server exposes a single MCP tool, search, which accepts a query string and returns relevant EIP content. Example usage in an MCP-compatible client:

You can use natural language prompts such as:

• "Search for EIPs related to CREATE2"
• "Find EIPs about contract deployment opcodes"
• "EIPs discussing CREATE2 implementation details"

The tool processes these queries and returns up to 5 relevant EIP document chunks, formatted with separators for clarity.

When searching for "CREATE2" using the search tool, the output might look like:

--------------------eip-1014.md--------------------
# EIP-1014: Skinny CREATE2
**Abstract**: This EIP adds a new opcode at 0xf5, CREATE2, which allows for deterministic address generation...
...
--------------------eip-1014.md--------------------
**Motivation**: The CREATE2 opcode enables predictable contract addresses, which is useful for...
...

License

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