FrankenClaw

Build your own agent tools in five minutes.

Drop a Python file in tools/, flip it on, done.

FrankenClaw

FrankenClaw isn't a pile of pre-built tools — it's the chassis. Auto-discovery, config-based enable/disable, and a single shared keys file are the whole product. Write one async function with a docstring, add it to your config, and your agent has a new tool. No server edits, no registration. It ships with one example tool you copy. You pick the models. You control the keys.

⚠ Important: This project is not affiliated with any cryptocurrency, token, or investment scheme. FrankenClaw is an open-source MCP tool chassis for AI agents, built by Project Sparks. If anyone offers you a “FrankenClaw token” — it’s a scam.

One function. One docstring. Done.

A FrankenTool is a single async function. The docstring is the description the model reads — so write it for the model. Save the file, flip it on, restart.

# tools/weather.py
import httpx

async def get_weather(city: str) -> str:
    """
    Get the current weather for a city.

    Args:
        city: City name, e.g. "Half Moon Bay".

    Returns:
        JSON with temperature and conditions.
    """
    async with httpx.AsyncClient() as client:
        r = await client.get(f"https://wttr.in/{city}?format=j1")
        return r.text

Turn it on in ~/.frankenclaw/config.json — that’s the entire workflow:

{ "enabled_tools": ["web_scrape", "get_weather"] }

Off by default. Zero tokens until you say so.

Every MCP tool definition costs your agent 500–1000 tokens on every call — used or not. A box of fifteen tools you mostly don’t need is a tax on every turn.

Off by Default
A tool you haven’t enabled is discovered but never registered. Zero tokens in tools/list.
Auto-Discovery
Drop a .py file in tools/. It appears as a candidate on next start. No registration.
One Keys File
API keys live in one place, read at runtime. A tool grabs its key with get_key().
Graceful Skip
A tool that fails to load is logged and skipped — the rest of the server still comes up.

FrankenClaw ships with one example tool — web_scrape (clean-markdown page scraping via Firecrawl) — as the template you copy. The former 15-tool bundle (search, vision, browser, Shopify, NotebookLM, Google Drive) is preserved on the archive/v0.3-bundled-tools branch — nothing deleted. Drop any of those files into tools/ and enable them to get them back.

Pure function. No opinions.

FrankenClaw does one thing: execute tool calls and return results. Everything else belongs to the agent.

No Memory
Your agent has Mnemo Cortex for that.
No Models
Your agent picks the models. Use any endpoint.
No Context
Your agent framework manages the conversation.
Just Tools
Request in, result out. The agent IS the agent.

One chassis. Purpose-built servers on top.

FrankenClaw is the open-source foundation. We also build focused MCP servers for specific jobs — same standard transport, same off-by-default discipline.

FrankenClaw
The open-source chassis. Build and run your own MCP tools. MIT-licensed, host-agnostic, available now.
Available
Shopify MCP
A purpose-built MCP server for Shopify merchants — the first focused server in the family. In development.
Coming Soon

Clone, install, point your agent at it.

git clone https://github.com/GuyMannDude/frankenclaw.git
cd frankenclaw
pip install -r requirements.txt

FrankenClaw speaks standard MCP over stdio. Connect it to any MCP host with one config block:

{
  "mcpServers": {
    "frankenclaw": {
      "command": "python3",
      "args": ["/ABSOLUTE/PATH/TO/frankenclaw/server.py"]
    }
  }
}

On Claude Code it’s one line: claude mcp add frankenclaw -- python3 /path/to/frankenclaw/server.py. A fresh clone ships with web_scrape enabled, so you have a working tool out of the box. Keys live at ~/.frankenclaw/keys.json; tool settings (including enabled_tools) at ~/.frankenclaw/config.json, auto-created on first run.

Works with
🦙 Any MCP Host

Claude Desktop, Claude Code, OpenClaw, Hermes Agent, LM Studio, AnythingLLM, Open WebUI, llama.cpp, Ollama, LobeChat, Jan — one config block each.

Linux/macOS native. On Windows, run inside WSL2 if a tool you add has system dependencies.

Setup Guides →

Part of the family.

Open-source AI infrastructure, built in Half Moon Bay.