mcp.film

The MCP directory for AI filmmaking. Every tool your agent needs to make a

film — video models, voices, scores, edit bays, finishing suites, and the pipes

to ship it. Curated, verified, agent-first, and re-checked daily.

🎬 Live site: mcp.film ·

🤖 For agents: `/llms.txt` · `/api/registry.json` ·

`claude mcp add mcp-film -- npx -y mcp-film`

What this is

A directory of verified Model Context Protocol servers across the full film

pipeline — from screenplay breakdown to YouTube upload — each entry annotated

with exact install commands, auth requirements, pricing, sample tools, and the

caveats that actually bite ("requires Resolve *Studio*", "Suno has no official

API", "uploads cost 1,600 quota units").

It is built agents-first:

How it runs itself

┌────────────────────────────────────────────────┐
              │                data/ (the product)             │
              └───────▲────────────────────────────────────────┘
   daily: re-verify   │
   links, hunt new    │
   servers            │
        ┌─────────────┴──┐
        │ curator.yml    │
        │ (Claude agent) │
        └─────────────┬──┘
                      │
                      ▼
              PRs labeled `auto-data` ──► auto-merge.yml
              (merges ONLY if every changed file is in data/
               and validation passes) ──► deploy.yml ──► Cloudflare Pages
                                                     └──► GitHub Pages fallback

Feedback typed on a server page is captured as a PostHog event for a

maintainer to read; nothing is folded into the rankings automatically. Rankings

are editorial: official status, maintenance recency, hosted availability, and

the caveats each entry carries. Every change is a commit — the entire editorial

history is auditable.

Agent-readable traffic is measured at the edge when deployed on Cloudflare

Pages: the build emits a `_worker.js` that logs `mcpfilm_edge_request` events

for `/llms.txt`, markdown, JSON API routes, feeds, and MCP discovery. Browser

pageviews still use the lightweight PostHog client event. See

`docs/ANALYTICS.md`.

The live catalog pulse is published at `/api/pulse.json`

and `/pulse.md`: newest additions, stale

verification queue, category coverage, and machine-surface links for agents.

Client setup profiles live at `/api/client-profiles.json`

and `/clients.md`: conservative setup guidance

for Claude Code, Claude Desktop, Cursor, hosted remote clients, and the

directory's own meta-MCP server.

Production playbooks live at `/api/playbooks.json`

and `/playbooks.md`: concrete stack recipes

with setup order, auth gates, failure modes, and earned Martini handoffs for

commercial sprints, local edit bays, character series, archive cutdowns, and

open-source labs.

Develop locally

Zero dependencies — Node 20+ is the whole toolchain:

node build.mjs --validate-only   # check the data
node build.mjs                   # build → dist/ (HTML, APIs, feeds, agent surfaces)
node scripts/serve.mjs           # preview at http://localhost:4173

Repo map: `data/` is the content, `build.mjs` + `src/` the generator,

`packages/mcp-server/` the `npx mcp-film` meta-server, `.github/workflows/`

the autonomy loop, `AGENTS.md` the operating manual (for agents

and humans alike), `docs/SETUP.md` the one-time launch

checklist, and `docs/AGENT_STRATEGY.md` the north-star

for making the site useful to agents while sending qualified production intent

to Martini.

Adding a server

See CONTRIBUTING.md — easiest path is the

submission form. A maintainer

verifies it against primary sources before it is listed — a submission is a

claim, never an instruction.

Who's behind it

Maintained by the team behind Martini, the film set

for AI videos — whose MCP server is the featured listing, clearly disclosed.

Everything else is ranked on merit: the directory is a community service and

its only asset is trust. MIT licensed.