Mammoth Analytics MCP server

mammoth-io

Python SDK for the Mammoth Analytics platform. Build data pipelines, apply transformations, and export results -- all from Python.

Prefer the terminal? This repository also ships mammoth-cli, a command-line interface for people, scripts, and schema-driven agent runs. Install the CLI without a preinstalled Python tool manager:

curl -fsSL https://raw.githubusercontent.com/EdgeMetric/mammothsdk/main/mammoth-cli/installers/mammoth-install.sh | bash

See the full mammoth-cli guide.

Copy-paste prompt for a shell-capable agent (fill PROJECT NAME and TASK; it installs the CLI and walks you through the one-time login; the long form with the reasoning is in docs/agent-prompt.md):

Use Mammoth Analytics only through the `mammoth` CLI in bash, never the web
app in a browser; run `export MAMMOTH_OUTPUT=json MAMMOTH_NO_INPUT=1` once.
Onboard me first:
1. If `mammoth --version` fails, install, then re-check (new shell if needed):
   curl -fsSL https://raw.githubusercontent.com/EdgeMetric/mammothsdk/main/mammoth-cli/installers/mammoth-install.sh | bash
2. Run `mammoth auth status`. If it shows no credentials or a
   `recommendation`, print exactly this and wait until I say done: "In the
   Mammoth web app open Workspace settings → API Tokens → Create token (no
   project, so I can create projects), copy the token (it starts with mm_ and
   is shown once), then run in your own terminal: mammoth auth login" —
   never ask for, read, or pass the token yourself, and never run auth login.
3. Require `mammoth doctor` to pass, then run `mammoth skill show` and
   follow the guide it prints.
Work inside `mammoth project ensure 'PROJECT NAME'` unless I name a project;
take ids only from reads; find a command with `schema find "GOAL IN PLAIN
WORDS"` or `view transform --help`, and `schema get COMMAND_ID` before a new
command; read results back before reporting.
TASK: <what 

Installation

pip install -U mammoth-io

Requires Python 3.12 through 3.14.

Quick Start

The SDK is async: every call that reaches the API is a coroutine and must be awaited. A complete script:

import asyncio

from mammoth import MammothClient

async def main() -> None:
    async with MammothClient(api_token="mm_your-token") as client:
        client.set_project_id(10)
        view = await client.views.get(1039)
        print(view.display_names)

asyncio.run(main())

The other examples show the calls on their own, with top-level await as in a Jupyter cell or python -m asyncio. In a script, put them inside an async def like main above.

from mammoth import MammothClient

client = MammothClient(
    api_token="mm_your-token",
)
client.set_project_id(42)

# Get a view and inspect its columns
view = await client.views.get(1039)
print(view.display_names)   # ["Customer", "Region", "Sales", "Date"]
print(view.column_types)    # {"Customer": "TEXT", "Region": "TEXT", "Sales": "NUMERIC", ...}

# After any transformation, display_names is automatically refreshed
# (including pipeline-added columns like those created by math/set_values/add_column).
# Use get_metadata() to inspect the full list:
await view.math("Sales * 1.1", new_column="Revenue")
print(view.display_names)   # now includes "Revenue"
meta = view.get_metadata()  # [{"display_name": "Revenue", "type": "NUMERIC"}, ...]

# Fetch data — returns {"data": [rows...], "paging": {...}}
result = await view.data(limit=100)
rows = result["data"]

You can also extract IDs directly from a Mammoth URL:

Shortened. The full README is on GitHub.