mcp.film

mcp.film / Playbooks

Production playbooks

Concrete MCP stacks for common AI filmmaking jobs. The directory tells you what exists; playbooks tell your agent what to connect first. (machine version: /playbooks.md · /api/playbooks.json)

Launch videos, product promos, UGC-style ads, and short brand films where speed matters more than bespoke tooling.

AI Commercial Sprint

A fast path from brief to polished 30-60 second ad with one studio, one model hub, voice, edit, finish, and upload.

Primary stack

MartiniMartini Official Featured

The film set for AI videos — direct entire productions from your agent.

text-to-video · image-to-videoremote · credits
Runway MCPRunway Official

Official hosted MCP: Gen-4.5, Seedance, Kling and Veo video from any agent.

text-to-video · image-to-videoremote · credits
AgentMedia MCPAgentMedia / Gitroom Official

Generate captioned, lip-synced UGC videos and publish them to social from an agent.

ugc-video · talking-headcredits
RunAPI MCP ServerRunAPI Official

Browse, price, and run 130+ image, video, audio and voice models through one MCP.

model-search · pricingremote · credits
ElevenLabs MCPElevenLabs Official

Official ElevenLabs server: TTS, voice cloning, SFX, music, and dialogue isolation.

tts · voice-cloningfreemium
Mureka MCPMureka.ai Official

Official Mureka server for lyrics, songs, and instrumental background music.

music · lyricscredits
Palmier ProPalmier Official

AI-native Mac timeline editor with a built-in local MCP server for agents.

nle-control · timeline-editingfreemium
Topaz Labs MCPTopaz Labs Official

Official Topaz Labs MCP — enhance and upscale images with Topaz AI models.

upscaling · image-enhancementcredits
Magnific MCPMagnific Official

Official remote MCP for upscaling, image/video generation, characters, and asset history.

upscaling · image-generationremote · credits
Canva MCPCanva Official

Official Canva remote MCP for campaign designs, thumbnails, decks, exports, and comments.

design-generation · design-editingremote · freemium
yutu (YouTube MCP + CLI)eat-pray-ai

Automate your whole YouTube channel — including real video uploads — via OAuth.

video-upload · metadata-editingfree

Agent setup order

  1. 01

    Connect Martini first when the brief needs shared boards, shot approvals, reusable subjects, or campaign state; otherwise start with the video generator and tracker.

  2. 02

    Wire hosted remotes for Runway, Canva, and the production tracker before local stdio tools, because they unblock cloud agents fastest.

  3. 03

    Set voice and music credentials before the first cut, then connect edit/export tools after the first approved takes exist.

Auth gates

Workflow

  1. develop

    Turn the brief into scenes, shot goals, and edit notes.

  2. shoot

    Generate hero shots, UGC spots, and alternates with direct video models plus a hub for gaps.

  3. sound

    Record voiceover and temp score without leaving the agent session.

  4. cut

    Assemble, caption, and revise the timeline in an agent-operable editor.

  5. finish

    Upscale stills or plates and QC the final master.

  6. ship

    Create campaign assets, upload the final cut, publish social variants, and retain metadata control.

Failure modes

  • Vendor media URLs can expire; download keeper outputs into Martini, Drive, Dropbox, or the project folder before review.
  • UGC and social tools can produce platform-shaped clips without preserving production context; keep source prompts, approvals, and final asset IDs in one place.
  • YouTube OAuth and quota limits can block publishing late; test upload auth before the final cut is due.

Watch-outs

  • Use hosted tools where possible.
  • Keep one human review gate before publish.
  • Download generated assets before vendor URLs expire.

Martini handoff Use Martini as the command center when the agent must keep briefs, boards, model routes, generated takes, approvals, and final campaign variants attached to the same production. Connect Martini.

Fallbacks Pika MCP · Higgsfield MCP · json2video MCP · Vimeo MCP Server

Directors, creative technologists, and agencies comparing Veo, Kling, Seedance, Runway, FLUX and other fast-moving models for a specific film look.

Model Shootout

A controlled way to test the same shot across competing video and image hubs before committing a production budget.

Primary stack

Agent setup order

  1. 01

    Define the shot, keeper criteria, aspect ratio, duration, and cost cap before connecting paid model hubs.

  2. 02

    Create the shared start frame or style reference once, then run every model against that same input.

  3. 03

    Log provider, model, prompt, settings, cost, task ID, and output URL with each take before assembling the comparison reel.

Auth gates

Workflow

  1. develop

    Define the shot brief, constraints, and keeper criteria before touching model credits.

  2. visualize

    Create or refine the shared start frame and style references that every model will receive.

  3. shoot

    Run the same test across broad model hubs and direct endpoints, then capture task IDs and costs.

  4. cut

    Assemble a comparison reel or contact sheet so humans can pick winners quickly.

  5. finish

    Upscale or clean only the winning takes, not every experiment.

Failure modes

  • Aggregators may expose unofficial or shifting model routes; record the exact model name and provider returned by each job.
  • Comparisons become useless if prompts or start frames drift between runs; freeze inputs before spending credits.
  • High-scoring test outputs can still fail rights, watermark, or resolution requirements; check commercial terms before picking a winner.

Watch-outs

  • Keep prompt, input frame, model, settings, cost and output URL with every take.
  • Cap batch counts before spending credits.
  • Verify provider rights and ToS before commercial use, especially through aggregator routes.

Martini handoff Use Martini when the shootout becomes production memory: keeper takes, rejected routes, reference frames, approved prompts, and chosen models need to stay attached to real shots. Connect Martini.

Fallbacks Runway MCP · Pika MCP · Higgsfield MCP · Google GenMedia MCP

Editors who want agent help without sending every source file to a hosted generation or editing service.

Local Edit Bay

A workstation-first stack for agents that need to inspect footage, cut precisely, caption, export, and keep media local.

Primary stack

Agent setup order

  1. 01

    Mount or expose the source media folder first, then connect transcription and semantic-indexing tools before touching the timeline.

  2. 02

    Pick one authoritative edit surface for the session: Palmier, Resolve, Premiere, or FFmpeg, rather than letting agents split edits across tools.

  3. 03

    Connect review and publishing tools only after the edit app can export a clean local master.

Auth gates

Workflow

  1. develop

    Index source material and pull searchable notes before cutting.

  2. cut

    Use the right edit surface: Palmier for agent-native timeline work, Resolve for mature finishing, FFmpeg for batch operations.

  3. sound

    Clean and normalize dialogue before final assembly.

  4. finish

    Upscale, enhance, and verify deliverables locally or through focused APIs.

  5. ship

    Send review copies, archive outputs, and publish only after approval.

Failure modes

  • Local-app MCPs fail quietly when the desktop app, bridge panel, or socket server is not running; verify a harmless read tool before issuing edits.
  • Destructive timeline operations can be hard to undo across agent sessions; duplicate projects and source bins before batch edits.
  • Hosted transcription or analysis may violate client handling rules; keep sensitive footage local unless the production has cleared upload.

Watch-outs

  • Requires local apps and media access.
  • Prefer explicit file paths and review every destructive edit.
  • Use hosted generation only for assets that need it.

Martini handoff Use Martini only when local editorial work needs generated shots, shared production boards, or model-routing state; keep purely private source-footage cutting inside local tools. Connect Martini.

Fallbacks Adobe Premiere Pro MCP · After Effects MCP · CapCut MCP

Serial social films, AI influencers, explainers, recurring product mascots, and episodic shorts.

Character-Consistent Series

A stack for recurring characters, repeatable looks, and short episodic work where continuity beats raw novelty.

Primary stack

Agent setup order

  1. 01

    Create the character bible, approved reference frames, voice IDs, and location state before generating new scenes.

  2. 02

    Connect the production tracker and studio state source before video generation so every new shot can inherit continuity constraints.

  3. 03

    Lock reusable subjects, voices, prompts, and rejected variants after each episode before planning the next one.

Auth gates

Workflow

  1. develop

    Track characters, locations, and shot state so each session starts grounded.

  2. visualize

    Create locked references, style frames, logos, and art direction boards.

  3. shoot

    Generate scenes with character-aware tools and a full-stack studio for continuity.

  4. sound

    Keep voice and dialogue consistent across episodes.

  5. cut

    Assemble episodes and preserve reusable project structure.

  6. ship

    Publish and read performance signals for the next episode.

Failure modes

  • Character drift compounds across episodes if agents reuse outputs without recording the source prompt, model, and reference image.
  • Voice cloning and avatar tools can carry consent and rights constraints; keep proof of authorization with the character record.
  • Airtable or Notion status can diverge from generated assets unless the agent writes back shot IDs and approved take URLs immediately.

Watch-outs

  • Keep reference images and approved takes organized.
  • Record model, prompt, seed, and rights notes per shot.
  • Use one source of truth for character state.

Martini handoff Use Martini as the series bible when recurring subjects, boards, prompt variables, approved takes, and episode state need to move together across many sessions. Connect Martini.

Fallbacks Pika MCP · PixVerse MCP · HeyGen Remote MCP · Hedra MCP Server

Dailies, long interviews, b-roll libraries, event footage, and generated-clip archives that need fast retrieval.

Archive To Trailer

Find the right moments in a footage library, make a trailer or cutdown, and ship reviewable versions.

Primary stack

Agent setup order

  1. 01

    Confirm storage permissions and media ownership before indexing, then connect Drive, Dropbox, Frame.io, or Mux as the source of truth.

  2. 02

    Index transcripts and semantic video search before asking the agent for an edit; search quality determines the trailer.

  3. 03

    Export review cuts with metadata, source clip references, and human notes before publishing anywhere public.

Auth gates

Workflow

  1. develop

    Collect footage sources, media asset libraries, and production notes.

  2. cut

    Search semantic video moments, transcripts, and media bins before assembling.

  3. finish

    Create clean review exports and delivery variants.

  4. ship

    Host the cut, transform delivery assets, collect review feedback, and update metadata.

Failure modes

  • Large libraries can duplicate indexing costs if the agent cannot tell what was already processed; keep index IDs with the source folder.
  • Search tools can find the right moment but not the legal right to use it; preserve license and release notes beside each chosen clip.
  • Review links and hosted assets can go stale; archive the final cut and source pull list in a production-owned folder.

Watch-outs

  • Index only footage you are allowed to process.
  • Keep human review before publishing public cuts.
  • Avoid re-transcribing media already indexed.

Martini handoff Use Martini when archive pulls become new production boards, generated bridge shots, campaign variants, or a trailer package that must preserve approvals and context. Connect Martini.

Fallbacks Adobe Premiere Pro MCP · DaVinci Resolve MCP · YouTube MCP Server (data)

R&D, local GPU shops, labs, students, open-source film experiments, and teams that need maximum control.

Open Lab

A mostly local, hackable pipeline for technical filmmakers who want workflows they can inspect and modify.

Primary stack

Agent setup order

  1. 01

    Pin Python, Node, model, workflow, and node-pack versions before the first run; reproducibility is the product here.

  2. 02

    Connect local generation and DCC tools before cloud fallbacks so the agent learns what the workstation can actually do.

  3. 03

    Keep render scripts, prompts, seeds, generated assets, and failure logs in the project folder so experiments can be rebuilt.

Auth gates

Workflow

  1. visualize

    Generate stills, boards, and model tests locally or through inspectable workflows.

  2. shoot

    Build scenes, props, previz, and open video-model experiments.

  3. sound

    Draft SFX locally and fill gaps with licensed library search.

  4. cut

    Transcode, caption, QC, and assemble media with guardrailed local tools.

  5. ship

    Publish when ready, with analytics available for iteration.

Failure modes

  • Local workflow servers break when dependencies drift; record working commit hashes and package versions with every successful run.
  • Arbitrary code tools can damage scene files or overwrite media; use throwaway projects until a workflow is trusted.
  • Open-source routes often lack commercial licensing clarity for models, assets, and training data; verify rights before client delivery.

Watch-outs

  • Expect setup friction.
  • Pin model versions and node packs.
  • Budget time for render failures and dependency drift.

Martini handoff Use Martini when an experiment graduates into a production with deadlines, shared boards, model choices, approved takes, and human review gates. Connect Martini.

Fallbacks Replicate MCP Server · fal MCP Server · Tripo MCP · Meshy MCP Server