Higgsfield Director MCP

Higgsfield Director MCP

Transforms Claude into a complete faceless-video studio for ten proven niches, producing timed storyboards, voiceover calls, captions, and execution manifests that plug directly into Higgsfield's official MCP server.

Category
Visit Server

README

Higgsfield Director MCP — the Faceless Channel Operating System

v4.0. Run this next to Higgsfield's official MCP server and Claude becomes a complete faceless-video studio: scriptwriter's editor, director, cinematographer, voice director, caption timer, budget controller and publishing assistant — for ten proven faceless niches.

One raw script in → a complete production out: narration-timed storyboard, the exact voiceover call, SRT captions, a cost comparison, an edit checklist, the publishing kit (titles/description/hashtags/thumbnail), and an execution manifest of real official Higgsfield MCP calls — price checks first, then character stills, then renders — that the agent pipes straight into the official server.

This layer is pure intelligence — it makes no network calls of its own. No account, no API key, zero dependencies, nothing leaves your machine. It thinks; the official server renders.

Higgsfield™ is a trademark of its respective owner. This is an independent, unofficial companion tool and is not affiliated with or endorsed by Higgsfield AI.


Why it pays for itself

AI video credits are the entire cost of running a faceless channel, and most of them are wasted on renders that get thrown away. The Director attacks that from five directions:

  1. Manifests that actually run. Every call uses the official server's real API shape: real catalog model IDs (veo3_1, kling3_0, minimax_hailuo, …), the real params envelope, durations snapped to what each backend truly accepts. No burned credits on malformed or improvised calls.
  2. Price checks before renders. Every manifest opens with get_cost: true preflight steps that return the true credit price of your shots — and a hard STOP until the budget is confirmed.
  3. Image-first economy (default). A keyframe is generated for every shot at image prices. You approve or re-roll keyframes for cents, then animate only the approved frames. Rerolling a failed idea costs a still, not a video.
  4. Audio off when narration covers it. Models that can render silent do, whenever your voiceover is the soundtrack anyway.
  5. A retention analyzer that runs before you spend. Weak hook, bloated beats, dead spots with nothing to shoot — caught and fixed while the script is still free to change.

Plus estimate_production_cost compares all three render strategies with the savings spelled out, and plan_budget fits your monthly credits to a posting schedule.

The niche matrix — 10 faceless formats, fully templated

Niche Aspect The look
historical_alt_documentary 16:9 oil-painting realism, god-rays, sweeping cranes
cosmic_horror_creepypasta 16:9 analog-horror VHS texture, chiaroscuro dread
dark_psychology_stoicism 9:16 monochrome marble macro, center-band captions
stickman_3d_documentary 16:9 matte-white 3D mannequin history dioramas
reddit_story 9:16 first-person POV realism, domestic unease
true_crime_documentary 16:9 90s news-archive grain, evidence macro, no faces
motivation_luxury 9:16 pre-dawn discipline b-roll, deep blacks
finance_money_psychology 9:16 editorial object metaphors on black glass
sleep_scary_stories 16:9 slow-burn cozy-dread ambience, 15s takes
biblical_mythology_stories 16:9 Caravaggio chiaroscuro, monumental scale

Every niche ships complete: recommended final and draft-tier models (real IDs), camera and shot-size rotations, safe zones, a retention rule, narration words-per-minute, voice direction, music notes, 10 tested hooks, CTA lines, title patterns, hashtags, and topic seeds that feed the 30-day series planner.

The tools (27)

The production line:

Tool What it gives you
produce_faceless_video The headline. Script + niche → narration clock, voiceover call, timed storyboard, SRT captions, cost comparison, edit checklist, publishing kit, and the full execution manifest. Optional save_as persists it to disk.
analyze_script_retention Score the script's hook, pacing and dead spots against the niche's rules — before credits are spent.
estimate_production_cost All-premium vs draft-video vs image-first, with savings spelled out.
plan_budget Monthly credits + upload schedule → the model tier that fits.
export_captions Narration-timed, valid .srt.
plan_shorts_cutdown 16:9 master → 9:16 shorts via the official reframe tool (a fraction of re-rendering).
plan_series 30–90 day calendar: topic + hook + title per episode.
package_for_platform 5 titles, paste-ready description, hashtags, thumbnail generate_image call.
channel_profile Set niche/voice/strategy once — saved to disk, every tool defaults to it.
list_productions / load_production Saved productions survive restarts; resume rendering any time.

The niche toolkit: get_viral_niches, generate_niche_storyboard, get_viral_hooks, apply_retention_framing, create_viral_short_clip, render_complete_script, assign_character_soul.

The director's toolkit: build_prompt, storyboard, apply_style / list_styles (12 cinematic looks), character_sheet, variations, suggest_preset, fix_prompt, negative_prompt.

Characters that never drift — and never get lost

Write recurring characters as @Tokens (@Hero draws her sword), register them once with assign_character_soul, and every script renders the same person. The roster is saved to disk and survives restarts — a series cast, not a session cast. Manifests automatically schedule a reference still per character (soul_2 with your Soul ID, or soul_cast from text) and wire its job_id into every shot's identity reference.

Built-in safety softening

Words that commonly trip generation filters (blood, gun, corpse, and a full true-crime/horror vocabulary) are swapped locally for visually equivalent phrasing before they ever reach the official server — the render queue keeps moving, and every substitution is reported.


New here? Read QUICKSTART.md

Zero-to-first-production in 5 minutes, no coding needed — Claude Desktop connection steps, paste-ready demo scripts in examples/scripts/, and a troubleshooting table.

Install

Requires Python 3.10+. Zero third-party dependencies.

pip install .
# or, from a checkout, run without installing:
python -m higgsfield_director --help

Use it with Claude Desktop (alongside the official server)

{
  "mcpServers": {
    "higgsfield": { "url": "https://mcp.higgsfield.ai" },
    "higgsfield-director": {
      "command": "python",
      "args": ["-m", "higgsfield_director", "serve"]
    }
  }
}

(Official-server entry shown for illustration — follow Higgsfield's own connection instructions for your client.)

Restart Claude Desktop, then just talk:

  • "Set my channel profile: niche true crime, image-first drafts."
  • "Produce this script as a stickman documentary." (paste the script — Claude returns the full production and runs the manifest on the official server, price checks first)
  • "Score this script before we render it."
  • "Plan my next 30 days of reddit stories."
  • "How many credits will this episode cost on each strategy?"

Use it as a CLI

higgsfield-director niches
higgsfield-director produce "Ten thousand workers haul stone at dawn." --as stickman
higgsfield-director analyze "My roommate was subletting my room." --as reddit
higgsfield-director cost "The lighthouse blinks wrong." --as creepypasta
higgsfield-director budget --credits 3000 --episodes 12 --aspect 16:9
higgsfield-director captions "Marcus wrote his journal at dawn." --as stoicism
higgsfield-director series --as motivation --days 30
higgsfield-director package --as "true crime" --topic "the payphone calls of 1994"
higgsfield-director profile --niche sleep --strategy image_first

Every command prints JSON, so it pipes cleanly into other tools.

Run the tests

python -m unittest discover -s tests

100+ offline tests, no network or API key required.

License

MIT — see LICENSE.

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured