Dawnwood for Agents — MCP server & CLI for AI drama production

THE POINT

Not another generate() button.

What is Dawnwood's agent access? An MCP server and CLI that expose the drama pipeline itself — runs, bible, scripts, set pack, renders — as agent tools, not just a text-to-video endpoint. As of August 2026, most video MCP servers hand your agent a fancier generate button: a prompt goes in, a clip comes out, and the assembling, continuity and story are still your problem. Here the agent is accountable for episodes: it drafts, revises and prices the season, and hands a human the console link at every approval gate.

HOW IT WORKS

Mint. Connect. Direct.

01 · MINT

One token from the console

Sign in, open the avatar menu → API tokens, and mint one. It expires in 90 days by default, is revocable anytime, is stored hashed on our side — and can never manage users, no matter who minted it.

02 · CONNECT

One command in your agent

claude mcp add dawnwood -- npx -y @dawnwood/cli mcp registers the server. Using Cursor, Claude Desktop or Codex? The setup guide has each client's config. Prefer raw HTTP? The same token works against the REST API — the route map lives at /llms.txt.

03 · DIRECT

You keep the judgment calls

The agent writes, revises and prices; at each gate — bible, scripts, set pack — it hands you a console deep link. Nothing renders that a human didn't approve.

QUICKSTART

Ninety seconds to first tool call.

# 1 — mint a token in the console (avatar menu → API tokens), then:
npx @dawnwood/cli auth login --token dw_pat_v1_…

# 2 — register the MCP server with your agent:
claude mcp add dawnwood -- npx -y @dawnwood/cli mcp

# 3 — open your agent and direct:
> take this premise, write the bible, and stop for my review

WHY IT MATTERS

Built for people whose editor is a terminal.

AUTOMATION STUDIOS

A season factory in your pipeline

Your orchestration already books flights and files PRs. Now it can take a premise to rendered episodes — with human sign-off gates exactly where your client expects them.

AGENT-NATIVE CREATORS

Direct in prose, not in panels

If you already live in Claude Code, the console becomes what it should be: the place you look at pictures and say yes — not the place you click through forms.

COST-CONSCIOUS OPS

Your agent's tokens are yours

The agent's reasoning runs on the subscription you already pay for it. Dawnwood charges the same plans and the same render credits — agent access adds nothing.

PRICING

It comes with your plan — not as an add-on.

$0 ON TOP OF ANY PLAN

Agent access is included in every plan, including Free. Plans are credit bundles; rendering burns 5 cr per output second (Super mode 10) and failed jobs are never charged — same numbers as the console, because it is the same pipeline. On Free, your agent can write the bible and scripts for nothing before a single render credit is spent.

Get your token

QUESTIONS

Asked before you asked.

Which agents work with Dawnwood?

Any MCP-compatible agent. Claude Code, Cursor and Codex all speak MCP; one command registers Dawnwood's stdio server. If your agent prefers raw HTTP, the same token works against the REST API — the machine-readable route map lives at /llms.txt.

What can the agent actually do?

Run the pipeline: create a season run, generate and revise the story bible, write every episode script, build the set pack, price and launch renders, and track them to finished episodes. Twenty-nine tools, mapping one-to-one onto the same pipeline the console drives.

What stays human?

Judgment. Every approval gate — bible, scripts, set pack — is designed for a human pass, and the agent hands you a console deep link at each one. Mastering and publishing to your connected channels also live in the console.

Does agent access cost extra?

No. It is included in every plan, including Free. Plans are credit bundles, rendering burns the same 5 credits per output second it always does (Super mode 10), and your agent's own reasoning runs on the subscription you already pay for it — Dawnwood never meters your agent's thinking.

How are tokens kept safe?

Personal access tokens expire after 90 days by default, are revocable at any time, are stored hashed server-side, and can never manage users regardless of who minted them. The CLI stores yours in a 0600 file and never prints it.

KEEP GOING

What your agent will be driving.

START NOW

Your agent is ready. The studio is too.

Mint a token, add one MCP server, and direct your first season from the terminal — the Free plan writes the bible and scripts before a single credit is spent.

Get your token Read the quickstart