Automation and agents

Labcoat treats automation as a public interface rather than a shell-screen scraping exercise.

JSON mode

Every command accepts --json and emits exactly one envelope on stdout. Logs and progress stay on stderr.

labcoat status --json
labcoat deploy my-token --dry-run --json

{
  "ok": false,
  "command": "deploy",
  "schema": "labcoat/v1/error",
  "error": {
    "code": "WALLET_MISSING",
    "message": "project wallet does not exist",
    "hint": "run `labcoat wallet init` first"
  }
}

An envelope being printed produces exit code 0, even when ok is false. Read the envelope rather than inferring application success from the process code.

MCP mode

labcoat mcp serve

The stdio MCP server exposes Labcoat Network, wallet, build, deployment, call, simulation, and trace tools using the same typed operations as the CLI.

Pin the Labcoat version used by an agent. The public website tracks current main; labcoat docs --llm is the installed-version authority.

Context endpoints

• /llms.txt is the concise index.
• /llms-full.txt contains the full public documentation.
• Every docs page has a sibling .md.txt URL.
• /reference/cli.json is the structured reference.
• /skill.md is the canonical workflow for coding agents.