Anaconda MCP

anaconda mcp is a CLI and server for exposing conda environment management tools to MCP-enabled AI coding assistants. It acts as a unified MCP endpoint, giving AI assistants like Claude, Cursor, and VS Code awareness of your conda environments, packages, and channel configurations.

📖 Full documentation: anaconda.com/docs · Development Guide

Prerequisites

• Conda (Miniconda or Anaconda Distribution)

Related MCP server: mcp_server_for_claudes_toolbox

Installation

conda create -n anaconda-mcp anaconda-mcp
conda activate anaconda-mcp

Installation via pip or uvx

Requires Python 3.10–3.14.

pip install anaconda-mcp

uvx anaconda-mcp@latest

Prerequisite: conda must be installed separately and discoverable on your PATH. pip and uvx do not provide conda. Without a working conda installation, the conda tools this server exposes will fail. Install Miniconda or Anaconda Distribution first, then follow the anaconda login and anaconda mcp terms accept steps below.

Important: The Anaconda MCP Server connects your conda environments to MCP-compatible AI assistants, enabling them to create, modify, and delete environments and packages on your machine. Install only if you trust the AI assistant you intend to connect and understand it can take real actions on your machine. By installing you acknowledge: (1) The AI assistant you connect to this MCP server is an independent third-party model, not a product or service of Anaconda. (2) Anaconda is NOT responsible for the actions the AI assistant directs within your environment, including unintended changes or deletions.

Anaconda Login

Authentication is required. The server will not start and tool calls will not succeed without a valid Anaconda login.

anaconda login

This opens a browser for OAuth login and stores the token in your system keyring. Subsequent starts use the stored token automatically.

Accept the Terms of Service

Anaconda MCP requires you to accept the Beta Terms before tool calls will succeed.

Interactive (recommended):

anaconda mcp terms accept

Non-interactive (CI / headless):

export ANACONDA_MCP_ACCEPTED_TERMS=true
export ANACONDA_MCP_ACCEPTED_TERMS_VERSION=2026-05-27

Setup

Configure your AI client to use Anaconda MCP:

anaconda mcp setup

This launches an interactive wizard that detects supported clients and writes the appropriate config. Supported clients: Claude Desktop, Claude Code, Cursor, Windsurf, VS Code, OpenCode, and Kilo Code.

To configure a specific client non-interactively:

anaconda mcp setup --client claude-code
anaconda mcp setup --client cursor --scope project
anaconda mcp setup --client kilo

Kilo Code Configuration

Anaconda MCP writes a plain kilo.json file (global ~/.config/kilo/kilo.json or project .kilo/kilo.json) which Kilo Code deep-merges with any existing kilo.jsonc. This means hand-written comments in kilo.jsonc are preserved across anaconda mcp setup runs. Important: if you hand-edit kilo.json itself with comments, anaconda mcp setup cannot parse it — it treats the file as empty and silently overwrites it, discarding your changes. Keep comments in kilo.jsonc instead.

Manual Client Configuration

If you prefer to configure your AI client manually, add an entry to your client's MCP config JSON. The anaconda mcp setup command resolves the Python path for you, but if writing by hand you'll need the full path to the environment's Python.

Example for Claude Code (.mcp.json):

{
  "mcpServers": {
    "anaconda-mcp": {
      "type": "stdio",
      "command": "/path/to/anaconda3/envs/anaconda-mcp/bin/python",
      "args": ["-m", "anaconda_mcp", "serve"],
      "env": {}
    }
  }
}

Optionally, you can pass authentication and TOS acceptance via the env dict instead of running anaconda login and anaconda mcp terms accept:

"env": {
  "ANACONDA_AUTH_API_KEY": "<your-api-key>",
  "ANACONDA_MCP_ACCEPTED_TERMS": "true",
  "ANACONDA_MCP_ACCEPTED_TERMS_VERSION": "2026-05-27"
}

API keys can be obtained from your Anaconda account settings.

⚠️ Each client has a different JSON schema. Check your client's MCP documentation carefully when writing configuration manually — field names and structure vary between Claude Code, Cursor, VS Code, and others.

Experimental: ana CLI

The ana CLI handles Anaconda MCP runtime installation and environment setup automatically, so you don't need to manage a dedicated anaconda-mcp conda environment path.

ana mcp serve installs and runs the managed Anaconda MCP runtime. It does not install Miniconda, Anaconda Distribution, or a user-facing conda command. If you want to activate or inspect environments from a terminal, notebook, or other local tooling, install and configure conda separately for that workflow.

Install ana:

curl -fsSL https://anaconda.sh/install.sh | sh

You still need to authenticate and accept TOS by running anaconda login and anaconda mcp terms accept beforehand.

When configuring your client manually with ana, use:

{
  "mcpServers": {
    "anaconda-mcp": {
      "type": "stdio",
      "command": "ana",
      "args": ["mcp", "serve"],
      "env": {}
    }
  }
}