⚡ Model Context Protocol · Open Source · Local-First

Turn any API
into an AI tool
in seconds.

MCPForge auto-generates Model Context Protocol (MCP) servers from OpenAPI specs. Connect Claude Desktop, Cursor, or any AI tool to any API — instantly.

Install MCPForge → View on GitHub

terminal — mcpforge

# Generate an MCP server from the GitHub API spec

$ mcpforge generate https://raw.githubusercontent.com/github/rest-api-description/main/descriptions/api.github.com/api.github.com.json

✓ Fetching OpenAPI spec...

✓ Parsed 847 endpoints — selecting top tools

✓ Enhancing descriptions with local AI...

✓ Generated 24 MCP tools

✓ Validated server — syntax OK

→ Written: github_mcp.py (32 tools, ready to run)

# Add to Claude Desktop — it now knows your API!

$ python github_mcp.py

✓ MCP server running — connected to Claude Desktop

How it works

Three steps. Zero boilerplate.

MCPForge handles all the plumbing so you can focus on building.

Point at your API spec

Pass an OpenAPI URL, Swagger file, or YAML spec. MCPForge fetches and parses it automatically — local or remote.

AI understands the API

Uses your local Ollama model to analyze endpoints, generate human-friendly descriptions, and identify the most useful tools to expose.

Run your MCP server

Get a single, self-contained Python file. Run it with python my_server.py. Point Claude Desktop or Cursor at it. Done.

Features

Everything you need to
build AI-ready tools

📄

OpenAPI → MCP in one command

Parse OpenAPI 3.0 and Swagger 2.0 specs. Every endpoint becomes an MCP tool with proper type hints, param validation, and docstrings.

🦙

AI-enhanced descriptions

Ollama models read your API docs and write developer-friendly tool descriptions. Better descriptions = better AI behavior. No cloud required.

🔑

Auth handled automatically

Detects Bearer tokens, API keys, and Basic auth. Injects them from environment variables so secrets never appear in your code.

✅

Built-in validation

Every generated server is syntax-checked and structure-validated before being written to disk. No broken files, ever.

🌐

Web UI included

Paste a spec URL, click generate, copy the result. The web UI lets non-CLI users generate MCP servers with zero friction.

📦

Ready-to-use examples

GitHub, Weather, and Calculator MCP servers included. Run mcpforge demo github to see a working MCP server in 10 seconds.

🔌

Works with every MCP client

Claude Desktop, Cursor, Zed, and any MCP-compatible tool. The output is standard MCP — no lock-in, no proprietary formats.

⚡

Zero config start

Works offline with local Ollama models. No API keys needed for the generator itself. Ollama enhancement is opt-in.

Built-in examples

See it in action

MCPForge ships with working examples you can run immediately.

🐙 GitHub API

Search repos, manage issues, view code — all from your AI assistant.

list_repos search_repos get_repo list_issues

🌤 Weather API

Get current weather and forecasts for any city — no API key needed.

get_weather get_forecast

🧮 Calculator

A simple math MCP server — perfect for testing your MCP client setup.

add subtract multiply calculate

bash

# Run the GitHub MCP example immediately

$ mcpforge demo github

# Or generate your own from any OpenAPI spec

$ mcpforge generate https://petstore.swagger.io/v2/swagger.json --output petstore_mcp.py

$ python petstore_mcp.py

Generated output

Clean, readable, production-ready

MCPForge generates Python you'd actually write yourself — with proper types, docstrings, and error handling.

          python — github_mcp.py (auto-generated)
        
#!/usr/bin/env python3
"""MCP server for GitHub API — auto-generated by MCPForge"""
import os
import httpx
from mcp.server.fastmcp import FastMCP
mcp = FastMCP("GitHub API")
BASE_URL = "https://api.github.com"
GITHUB_TOKEN = os.environ.get("GITHUB_TOKEN", "")
@mcp.tool()
def list_repos(owner: str, per_page: int = 30) -> dict:
    """List public repositories for a GitHub user."""
    headers = {"Authorization": f"Bearer {GITHUB_TOKEN}"} if GITHUB_TOKEN else {}
    resp = httpx.get(f"{BASE_URL}/users/{owner}/repos", params={"per_page": per_page}, headers=headers)
    resp.raise_for_status()
    return resp.json()
# ... 23 more tools generated ...
if __name__ == "__main__":
    mcp.run()

Installation

Up and running in 30 seconds

bash

# Install

$ pip install mcpforge

# Generate your first MCP server

$ mcpforge generate https://petstore.swagger.io/v2/swagger.json

# Generate + print Claude Desktop config in one step

$ mcpforge generate https://petstore.swagger.io/v2/swagger.json --claude-config

# Run it

$ python petstore_mcp.py

Add to Claude Desktop — use --claude-config to auto-generate, or manually edit claude_desktop_config.json:

json

{ "mcpServers": { "github": { "command": "python", "args": ["/path/to/github_mcp.py"], "env": { "GITHUB_TOKEN": "your-token-here" } } } }

Turn any API into an AI tool in seconds.

Three steps. Zero boilerplate.

Everything you need tobuild AI-ready tools

See it in action

Clean, readable, production-ready

Up and running in 30 seconds

Ready to forge your firstMCP server?

Turn any API
into an AI tool
in seconds.

Everything you need to
build AI-ready tools

Ready to forge your first
MCP server?