AI agents

Using Anesis with AI agents and CI

Anesis is scriptable end to end: non-interactive flags, JSON output, token auth, and an anesis mcp server that exposes the CLI to any MCP-capable agent.

ANESIS_TOKEN--yes--jsonanesis mcp

Authenticate with a token

Personal access tokens replace the browser login for CI and agents.

Create a token on the API tokens page. Set it as ANESIS_TOKEN and every authenticated command uses it — the plaintext token is shown only once, so store it securely.

# 1. Create a token at /account/tokens, then:
export ANESIS_TOKEN=anesis_pat_...

# 2. Every authenticated command now uses it — no browser login.
anesis template install react-vite-ts
anesis new my-app react-vite-ts --yes

Non-interactive flags

These make every command safe to run unattended.

--yes accepts all defaults and skips every confirmation prompt.
--input NAME=VALUE supplies an input non-interactively (repeatable).
--json prints machine-readable output on list, info, search, status, account, and info commands.
ANESIS_TOKEN authenticates the CLI in place of the browser login flow.

The MCP server

anesis mcp runs a Model Context Protocol stdio server over the CLI.

Run anesis mcp to start a stdio JSON-RPC server. Point any MCP-capable client (e.g. Claude Code) at it with this configuration:

{
  "mcpServers": {
    "anesis": {
      "command": "anesis",
      "args": ["mcp"],
      "env": { "ANESIS_TOKEN": "anesis_pat_..." }
    }
  }
}

Exposed tools

search_registry — search templates, addons and stacks.
get_manifest — full manifest for a template, addon or stack.
scaffold_project — create a project from a template.
apply_addon — run an addon command in a project.
apply_stack — scaffold a project from a stack (template + ordered addons).
project_status — the current project's template and applied addons.

CLI Commands